To compare the risk of contrast-associated acute kidney injury (CA-AKI) caused by iso-osmolarity contrast media (IOCM) and low-osmolarity contrast media (LOCM) in pediatric patients. A retrospective cohort exposed to contrast-enhanced CT was constructed based on the inpatients from Beijing Children’s Hospital (2015–2020). The patients were divided into IOCM and LOCM exposure. AKI was defined by the pediatric version of Kidney Disease: Improving Global Outcomes (KDIGO, pKDIGO) criteria. After adjusting baseline differences with inverse probability of treatment weighting (IPTW), log-binomial regression was conducted to estimate the risk ratios (RR) and 95

In the task of predicting pediatric reference intervals, indirect estimation methods are increasingly viewed as a promising complementary strategy for direct sampling, offering a way to gather insights without imposing the full burden of direct data collection on healthy children. However, existing indirect estimation methods rely solely on individual analytes for predicting their reference intervals, neglecting the impact of other biomarkers on the prediction task and consequently reducing the accuracy of reference interval predictions. In this paper, we propose an auto-weighted multi-dimensional biomarkers graph structure learning for deep clustering (AMGSL). To enhance the accuracy of reference interval estimation for target analyte using multiple biomarkers, our algorithm captures the influence of auxiliary biomarkers on the target analyte by designing a weight learning mechanism, even in the absence of biomarker correlation information. In the process of graph structure learning, we design a distance function that takes the obtained attention weights between biomarkers and the dataset size as internal variables, calculating the similarity between samples. In addition, we propose a hyperparameter optimization strategy based on internal metric indicators to achieve optimal model performance. Extensive experiments were conducted on five data subsets corresponding to three analytes derived from the clinical dataset; when the experimental results were compared with those of existing algorithms, they demonstrated the superiority of the proposed method. Furthermore, a comparison against medical expert knowledge has also validated the effectiveness of the proposed weight learning mechanism.

Abstract While ETV6 :: RUNX1 -positive acute lymphoblastic leukemia (ALL) is generally associated with favorable outcomes, a subset of patients experience relapse despite receiving standard therapy, underscoring the need for early biomarkers to identify high-risk subgroups. In this retrospective study of 345 consecutive pediatric patients with ETV6 :: RUNX1 -positive ALL treated in accordance with the protocol of the Chinese Children’s Leukemia Group (CCLG-ALL 2008), CD33 expression (CD33 + ) on leukemic blasts was detected in 55.9% of patients and was significantly associated with minimal residual disease (MRD) positivity on day 15 (D15-MRD-positive) (65.2% vs. 42.0%, P < 0.001). Multivariate analysis revealed CD33 + status as an independent risk factor for D15-MRD positivity (odds ratio (OR) = 2.66, 95% confidence interval (CI) 1.66–4.26). Among CD33 + patients, those who were D15-MRD-positive had significantly inferior 5-year and 10-year event-free survival (EFS) relative to those who were D15-MRD-negative (5-year EFS: 91.9% ± 2.4% vs. 100%, P = 0.027; 10-year EFS: 89.2% ± 3.0% vs. 100%, P = 0.014). This prognostic association was not observed in CD33-negative patients. In conclusion, the combination of CD33 + and D15-MRD-positivity may identify a distinct high-risk subgroup within ETV6 :: RUNX1 -positive ALL. Early intervention, potentially including CD33-directed therapy, may represent a promising strategy to improve outcomes in this subgroup, although further validation is warranted.

BACKGROUND:While artificial intelligence's (AI's) transformative potential in health care is widely acknowledged, its application in highly sensitive, humanistic domains like pediatric palliative care (PPC) remains largely unexplored. OBJECTIVE:This study aims to explore the attitudes and needs of health care providers on the PPC assisted by AI, with the goal of informing future development and implementation of AI systems in this field. METHODS:This was an explanatory sequential mixed methods study consisting of a nationwide cross-sectional questionnaire survey (March-April 2025) followed by qualitative semistructured interviews (August-October 2025). The quantitative study aimed to investigate PPC health care providers' experiences, attitudes, and needs for the application of AI. Participants included team members of all recognized PPC teams in mainland China. The qualitative study aimed to explore in greater depth the potential future roles of AI in this field, as well as the features of an ideal AI-assisted tool for PPC. Potential interviewees were recruited from the pool of quantitative survey respondents. RESULTS:Among 352 survey respondents, most (n=205, 58.24%) reported moderate familiarity with AI, with large language models being the most commonly used (n=280, 79.55%). Among large language model users, over half (161/280, 57.50%) reported using them for clinical purposes. Attitudes were generally positive: 67.05% (236/352) believed AI's benefits would outweigh drawbacks, and 75% (264/352) considered its implementation feasible. The most desired applications were patient and family education (276/352, 78.41%) and symptom management (257/352, 73.01%). Interviews with 17 providers revealed three themes: (1) clinical roles and boundaries, (2) elements for clinical integration, and (3) challenges in development and deployment. CONCLUSIONS:This study reveals that PPC providers express positive attitudes and strong demand for AI-assisted clinical work. Furthermore, the research clarifies appropriate roles for AI, outlines elements for clinical integration, and highlights potential challenges in development and integration. This study provides evidence for the feasibility of AI application in PPC and offers guidance for the future development and deployment of AI tools.

Background Birth defects, which comprise a series of severe congenital abnormalities, impose a significant burden on society, families and individuals. Consequently, it is crucial to identify the underlying causes of birth defects and reduce their occurrence. Although an increasing number of risk factors for birth defects have been identified, few associations can be established as causal. Furthermore, the distribution of aetiology related to birth defects remains unclear. This study aims to analyse birth defect cases from the China Birth Cohort Study (CBCS) to elucidate the aetiological profile of these conditions.Methods A total of 3873 abnormal cases were recorded in the CBCS from November 2017 to August 2021. Abnormal fetuses (including both live births and foetal losses) were diagnosed by obstetricians, ultrasound specialists and geneticists based on prenatal screening and clinical examinations. The causes of birth defects were categorised into chromosomal anomalies, genetic anomalies, environmental exposures and twinning. Chromosomal and genetic anomalies were identified through genetic screening. Data on exposure, including the substances involved and the duration of exposure, were reviewed to determine whether environmental factors contributed to the birth defects.Results After excluding cases with minor malformations, a total of 2123 birth defect cases were reviewed. The most common birth defects among the included cases were congenital heart disease, polydactyly, trisomy 21 and cleft palate with cleft lip. Of these, only 22.4% (475/2123) had identifiable causes. Specifically, 415 cases were attributed to chromosomal anomalies, while 31 cases were diagnosed as monogenic disorders. Additionally, 23 cases were linked to environmental exposures, and 6 cases were associated with twinning. The proportions of birth defect cases with known causes were significantly higher in the spontaneous abortion group (12/27, 44.4%), the therapeutic abortion group (314/1044, 30.1%) and perinatal death group (13/36, 36.1%) compared with live births (136/1016, 13.4%).Conclusions Nearly 80% of birth defect cases in the CBCS lack a clear identifiable cause. Therefore, translating statistical associations between risk factors and birth defects into causal relationships is both necessary and important.

BACKGROUND AND OBJECTIVE:To systematically evaluate the performance of k-fold cross-validation and bootstrap-based optimism correction methods for internal validation of statistical and machine learning models. METHODS:A total of 239,415 inpatients were extracted from an open access database named Medical Information Mart for Intensive Care IV, of which 39,145 were randomly sampled as a predefined reference dataset. Among the remaining simulation dataset with 200,000 inpatients, training sets with sample sizes ranging from 595 to 5946 were randomly selected, and multiple prediction models were developed in each training set using various modeling strategies, including logistic regression, least absolute shrinkage and selection operator regression, Naive Bayes, Support Vector Machine, K-Nearest Neighbors, Light Gradient Boosting Machine, and Random Forest. The dependent variable of the model was acute kidney injury (AKI), a binary outcome with an incidence of 18.5%, and the independent variables included 22 common predictors of AKI. For each model, 2-fold, 5-fold, and 10-fold cross-validation were used for internal validation to calculate area under the receiver-operating characteristic curve (AUC), which is a common metric for quantifying the overall ability of a model to discriminate between positive or negative classifications. In addition, the Harrell, .632, and .632+ AUC estimators were calculated for internal validation based on bootstrapping. The above simulation process was repeated 1000 times to obtain 1000 estimates of AUC for each internal validation method of each model. The model performance was simultaneously evaluated in the reference dataset to obtain an empirical AUC (analogous to the "gold standard"). Then, by comparing the 1000 AUC estimates with the empirical AUC, the accuracy of internal validation methods for different models was assessed. RESULTS:For parametric models, the .632+ estimator provided the most accurate estimates of AUC, followed by 10-fold cross-validation with only slight bias. In contrast, for nonparametric models, all bootstrap-based optimism correction methods significantly overestimated AUC, and the overestimation was not reduced by increasing the sample size. Most strikingly, 10-fold cross-validation demonstrated stable and good performance across all scenarios considered, regardless of the modeling strategy or sample size. CONCLUSION:The performance of bootstrap-based optimism correction methods can be affected by model complexity, although the .632+ estimator performs best in parameter models based on small-sample training. In comparison, 10-fold cross-validation is more robust and easier to implement. Therefore, it is recommended to prioritize 10-fold cross-validation as the internal validation method for prediction models. PLAIN LANGUAGE SUMMARY:With the exponential growth of clinical prediction models, the methods for conducting internal validation of these models remain controversial. Both k-fold cross-validation and bootstrap-based optimism correction methods are recommended by guidance papers. However, the issue of whether they are applicable to all modeling strategies, especially machine learning algorithms, still lacks evidence. This study simulated various sample size scenarios based on real-world clinical data, and developed AKI prediction models based on parametric and nonparametric modeling strategies. Then, internal validation was performed for each model using different methods. The results showed that bootstrap-based optimism correction methods were suitable for parametric models. However, as the model complexity increased, the bias of bootstrap-based optimism correction methods increased accordingly. In contrast, 10-fold cross-validation performed well in all scenarios, regardless of the modeling strategy or sample size. Therefore, 10-fold cross-validation is recommended as a preferred method for internal validation of prediction models.

OBJECTIVE:To compare the perinatal outcomes between fresh and frozen embryo transfer strategies among ongoing pregnancies conceived by in vitro fertilization/intracytoplasmic sperm injection. DESIGN:A prospective cohort study. The study was conducted within the framework of a hypothetical randomized controlled trial to enhance the validity of the evidence obtained from observational data. SUBJECTS:From November 2017 to August 2021, 5,118 pregnant women who conceived by in vitro fertilization/intracytoplasmic sperm injection were recruited from 50 study sites in 17 provinces of China during their first trimester. EXPOSURE:Frozen vs. fresh embryo transfer. MAIN OUTCOME MEASURES:The primary outcome was perinatal complications, defined as any occurrence of perinatal death or birth defects. The secondary outcomes included preterm birth, small for gestational age, large for gestational age, low birth weight, and macrosomia. The safety outcomes were abortion, pregnancy-induced hypertension, gestational diabetes mellitus, and gestational thyroid dysfunction. RESULTS:A total of 2,856 pregnant women were included in the analysis, with the allocation ratio of 1:1. The perinatal complication rate of the frozen embryo transfer group (5.0%, 72/1,428) was similar to that of the fresh embryo transfer group (4.6%, 66/1,428), with the risk ratio of 1.09 (95% confidence interval, 0.79 to 1.51). Moreover, there was no significant difference in the risks of preterm birth, small for gestational age, large for gestational age, low birth weight, and macrosomia between the two groups. However, compared with fresh embryo transfer, frozen embryo transfer was associated with an increased risk of pregnancy-induced hypertension (risk ratio, 2.18; 95% confidence interval, 1.10 to 4.62). CONCLUSION:The risk of perinatal complications was similar between fresh and frozen embryo transfer strategies, whereas the risk of pregnancy-induced hypertension seemed to be higher for the frozen embryo transfer strategy among ongoing pregnancies. Therefore, the decision regarding fresh or frozen embryo transfer should be made with more caution, with careful consideration of the benefits and potential risks.

Acute kidney injury is associated with a prolonged hospital stay and high mortality for pediatric patients. The previous prediction models are based on a pre-defined time window which may affect its feasibility in clinical practice. This study aimed to develop and validate a real-time acute kidney injury risk prediction model for hospitalized pediatric patients. Based on a retrospective cohort composed of eligible pediatric patients hospitalized in Beijing Children’s Hospital and Chongqing Children’s Hospital, a machine learning model to predict acute kidney injury occurrence was developed and validated. The prediction model was established using a stacking technique to combine three base learners including XGBoost, LightGBM, and CatBoost. Particle swarm optimization algorithm was used to tune hyperparameters. Next, we assessed the performance of the prediction model using the area under the receiver-operating curve and the area under the precision–recall curve. A total of 26,671 patients were included (20,967 in the derivation set, 5242 in the internal validation set, and 462 in the external validation set) contributing to 36,828 hospitalizations. The new proposed model had excellent performance for predicting any acute kidney injury within 24 hours. Both the internal [area under the receiver-operating curve (AUROC) was 0.851, area under the precision–recall curve (AUPR) was 0.322] and external validation set (AUROC was 0.869, AUPR was 0.270) approved the model’s feasibility in predicting the risk of acute kidney injury. The established risk predicting model can be used for real-time prevention of acute kidney injury in hospitalized pediatric patients.

AIM:To develop a comprehensive decision-making checklist for paediatric advance care planning tailored to the needs of terminally ill children and their families. DESIGN:A Delphi Study. METHODS:Underpinned by Delphi methodology, a four-phase procedure was adopted: (1) drafting items by the working group, (2) refining items based on an experts' survey, (3) further refining based on the same experts, and (4) final adaptations and approval. This study was initiated by the Paediatric Palliative Care Subspecialty Group of the Paediatrics Society of the Chinese Medical Association. The process involved 60 healthcare providers (physicians, nurses, and social workers) from 14 paediatric palliative care teams. RESULTS:The developed checklist included 5 topics, 24 subtopics, and 45 items. Five topics were (1) medical and nursing decision-making, (2) social support planning, (3) psychological support planning, (4) spiritual support planning, and (5) posthumous affairs planning. This checklist addresses symptom management for terminally ill children, integrating support for their psychological, social, and spiritual well-being, and addresses the care needs of their family members. CONCLUSIONS:The study provided a paediatric advance care planning checklist derived from the expert consensus that includes key elements and items. IMPLICATIONS:This checklist provides healthcare providers with a structured framework to set paediatric advance care planning and ensure that all aspects of children's well-being and their families' needs are considered. This study also lays an evidence-based foundation for the design of related documents. IMPACT:This study developed a comprehensive paediatric advance care planning checklist with 5 topics, 24 subtopics, and 45 items. This study provides a comprehensive decision-making checklist for healthcare providers and families, ensuring that critical decisions are addressed timely. REPORTING METHOD:This article is presented in accordance with the CREDES guidelines. PATIENT OR PUBLIC CONTRIBUTION:Limited patient and public involvement was incorporated, focusing on reviewing the initial checklist draft.

How to assess the importance of predictors in systematic reviews (SR) of prediction models remains largely unknown. The commonly used indicators of importance for predictors in individual models include parameter estimates, information entropy, etc., but they cannot be quantitatively synthesized through meta-analysis. We explored the synthesis method of the importance indicators in a simulation study, which mainly solved the following four methodological issues: (1) whether to synthesize the original values of the importance indicators or the importance ranks; (2) whether to normalize the importance ranks to a same dimension; (3) whether and how to impute the missing values in importance ranks; and (4) whether to weight the importance indicators according to the sample size of the model during synthesis. Then we used an empirical SR to illustrate the feasibility and validity of the synthesis method. According to the simulation experiments, we found that ranking or normalizing the values of the importance indicators had little impact on the synthesis results, while imputation of missing values in the importance ranks had a great impact on the synthesis results due to the incorporation of variable frequency. Moreover, the results of means and weighted means of the importance indicators were similar. In consideration of accuracy and interpretability, synthesis of the normalized importance ranks by weighted mean was recommended. The synthesis method was used in the SR of prediction models for acute kidney injury. The importance assessment results were approved by experienced nephrologists, which further verified the reliability of the synthesis method. An importance assessment of predictors should be included in SR of prediction models, using the weighted mean of importance ranks normalized to a same dimension in different models.

This study aimed to adapt the existing Q-based equation for glomerular filtration rate (GFR) estimation using serum creatinine (SCr) to Chinese children by modification of certain parameters, and to establish the pediatric reference intervals (RIs) for estimated GFR in China accordingly. The Q-based equation had the form eGFR = M/(SCr/Q), where the parameter M was estimated from 328 children with measured GFR in Wuhan Children’s Hospital, and the parameter Q was fitted among 11,713 healthy children volunteers who had SCr recruited from 11 provinces of China. The Q-based equation was applied to 60,524 inpatients in Beijing Children’s Hospital to assess the impact of estimated GFR on the clinical diagnosis of acute kidney injury (AKI). Then, the Q-based equation was used to estimate GFR in a large representative population of healthy Chinese children for RI establishment. The parameter M was estimated separately in children aged less than 2 years (100.2) and 2 years or above (107.3). The parameter Q was modeled as a linear function of age in boys (19.5 + 3.2Age) and girls (23.6 + 2.2Age), respectively. The GFR estimated by the Q-based equation could well identify AKI, with a significantly higher risk of in-hospital mortality in AKI patients than non-AKI patients (OR, 5.69; 95

Backgrounds and objectives Accurate detection and staging of acute kidney injury (AKI) is important in clinical practice to aid in timely management. The main purpose of this study is to establish a pediatric version of KDIGO (pKDIGO) criteria for pediatric population. Methods The pKDIGO criteria defined AKI following the principles of KDIGO, in which the threshold of absolute increase in SCr or absolute decrease in eGFR to diagnose AKI has been revised to eliminate the impacts of age and gender of children. Then, AKI defined by pKDIGO were compared with that defined by KDIGO, mKDIGO, pROCK, and pRIFLE based on two retrospective cohorts in China: Beijing Children’s Hospital (BCH) cohort and Intensive Care Units of the Children's Hospital of Zhejiang University School of Medicine (ICU) cohort. The performance of different AKI definitions was compared based on the area under the receiver operating characteristic curves (AUCs) for predicting the in-hospital death. Results Total of 57 229 children in the BCH cohort and 8276 children in the ICU cohort were used to evaluate the performance of pKDIGO. In the BCH cohort, AUCs for predicting mortality by AKI defined based on pKDIGO (AUC=0.75, 0.72-0.78) were higher than that defined by other definitions. The risk of death increases with higher stage of AKI defined by pKDIGO. Similar results were also observed in the ICU cohort. Conclusions The pKDIGO criteria showed a better ability to identify AKI patients and predict in-hospital death in children, whatever in general wards or ICUs.

Background:Acute kidney injury (AKI) is common in hospitalized children. A post-AKI outcomes prediction model is important for the early detection of important clinical outcomes associated with AKI so that early management of pediatric AKI patients can be initiated. Methods:Three retrospective cohorts were set up based on two pediatric hospitals in China, in which 8205 children suffered AKI during hospitalization. Two clinical outcomes were evaluated, i.e. hospital mortality and dialysis within 28 days after AKI occurrence. A Genetic Algorithm was used for feature selection, and a Random Forest model was built to predict clinical outcomes. Subsequently, a temporal validation set and an external validation set were used to evaluate the performance of the prediction model. Finally, the stratification ability of the prediction model for the risk of mortality was compared with a commonly used mortality risk score, the pediatric critical illness score (PCIS). Results:The prediction model performed well for the prediction of hospital mortality with an area under the receiver operating curve (AUROC) of 0.854 [95% confidence interval (CI) 0.816-0.888], and the AUROC was >0.850 for both temporal and external validation. For the prediction of dialysis, the AUROC was 0.889 (95% CI 0.871-0.906). In addition, the AUROC of the prediction model for hospital mortality was superior to that of PCIS (P < .0001 in both temporal and external validation). Conclusions:The new proposed post-AKI outcomes prediction model shows potential applicability in clinical settings.

BACKGROUND:Pediatric palliative care (PPC) should be integrated throughout the disease trajectory and recognized as a fundamental component of pediatric healthcare systems. AIM:To describe the development level of PPC in mainland China and elucidate the changing characteristics in this field over the past five years. METHODS:This is a cross-sectional study initiated by the PPC subspecialty group of the Pediatrics Society of the Chinese Medical Association. Study participants included all PPC teams in mainland China. The questionnaire was structured into five main sections: team characteristics, personnel composition, service quality, service types and contents, and support needs. This study compared the level of PPC development in mainland China in 2025 with that of 2019, with the 2019 data sourced from literature published by our team. Descriptive statistics were used to analyze the data. RESULTS:There were 36 PPC teams, covering 16 out of 31 (51.61%) provinces and municipalities in mainland China. The median total number of team members was 12 (9, 16). From 2019 to 2025, the number of teams decreased from 45 to 36. However, the total number of team members increased from 300 to 513, demonstrating significant structural improvements in PPC teams. With the improved team structure, the content of palliative care services has significantly expanded. CONCLUSION:Over the past five years, the PPC system in mainland China has made improvements to its team structure and dynamics. Despite this progress, it continues to face interrelated challenges: severe resource shortages, inefficient resource utilization, and uneven geographical distribution of services.

Introduction:Accurate detection and staging of acute kidney injury (AKI) is important in clinical practice to aid timely management. The main purpose of this study is to establish a pediatric version of Kidney Disease: Improving Global Outcomes (KDIGO, pKDIGO) criteria for pediatric population. Methods:The pKDIGO criteria defined AKI following the principles of KDIGO, in which the threshold of absolute increase in serum creatinine (SCr) or absolute decrease in estimated glomerular filtration rate (GFR, eGFR) to diagnose AKI has been revised to eliminate the impacts of age and sex of children. Then, AKI defined by pKDIGO were compared with that defined by KDIGO, modified KDIGO (mKDIGO), pediatric reference change value optimized for AKI in children (pROCK), and pediatric Risk for renal dysfunction, Injury to the kidney, Failure of kidney function, Loss of kidney function, and End-stage renal disease (RIFLE, pRIFLE) based on 2 retrospective cohorts in China: Beijing Children's Hospital (BCH) cohort and intensive care units (ICUs) of the Children's Hospital of Zhejiang University School of Medicine (ICU) cohort. The performance of different AKI definitions was compared based on the area under the receiver operating characteristic curves (AUCs) for predicting the in-hospital death. Results:Total of 57,229 children in the BCH cohort and 8276 children in the ICU cohort were used to evaluate the performance of pKDIGO. In the BCH cohort, AUCs for predicting mortality by AKI defined based on pKDIGO (AUC = 0.75, 0.72-0.78) were higher than that defined by other definitions. The risk of death increases with higher stage of AKI defined by pKDIGO. Similar results were observed in the ICU cohort. Conclusion:The pKDIGO criteria showed a better ability to identify patients with AKI and predict in-hospital death in children, both in general wards and ICUs.

BACKGROUND:Drug-induced thrombocytopenia (DITP) often occurs in patients during clinical treatment. However, clinicians usually fail to distinguish which drugs can be plausible culprits accurately. We aimed to develop a large comprehensive drug benchmark database with DITP toxicity using the recommended method by FDA. RESEARCH DESIGN AND METHODS:We collected information from six databases that involved drug labeling information, literature, safety signal mining and laboratory testing to generate the annotated drug list with DITP toxicity. Then, we descripted the DITP positive-negative distribution based on the Anatomical Therapeutic Chemical (ATC) coding system; hotspot analysis was conducted to identify therapeutic categories of drugs within each organ system that warrant attention regarding DITP. RESULTS:The DITPst database comprised 1,765 drugs, of which 858 were DITP-positives, whereas 907 were negatives. The investigation of distribution across various therapeutic categories revealed the most frequent DITP-positive categories were immunostimulants (10/11), anti-inflammatory, and antirheumatic products (28/32), and antibacterials for systemic use (102/121). On the contrary, the least frequent DITP-positive therapeutic categories were diagnostic radiopharmaceuticals (12/12), pituitary and hypothalamic hormones and analogues (17/18), and drugs for constipation (16/17). CONCLUSIONS:We consider the DITPst benchmark database to be an invaluable resource for the community to improve DITP safety research and drug development.

ABSTRACT Importance Although macrolides combined with glucocorticoid therapy have demonstrated efficacy in preventing long‐term pulmonary lesions of severe Mycoplasma pneumoniae pneumonia (MPP), evidence regarding glucocorticoid dose is lacking. Objective To evaluate the effects of low‐ and high‐dose methylprednisolone on the risk of long‐term pulmonary lesions for children with severe MPP when combined with azithromycin. Methods This randomized, parallel‐controlled, multicenter clinical trial was conducted in mainland China and enrolled pediatric patients hospitalized with severe MPP. A total of 424 enrolled patients were randomized (allocation ratio of 1:1) to azithromycin combined with either a low‐dose [2 mg/(kg·d)] or a high‐dose [10 mg/(kg·d)] methylprednisolone treatment for 3 d followed by tapering over 12 d. The primary outcome was the incidence of composite adverse outcomes, including atelectasis, bronchiectasis, or bronchiolitis obliterans 6 months after treatment. Results A total of 118 (27.8%) developed adverse pulmonary lesions at 6 months after treatment; 66 of 211 (31.3%) in the high‐dose methylprednisolone group and 52 of 213 (24.4%) in the low‐dose group, respectively. The risk ratio of long‐term pulmonary lesions in a high‐dose group to those in a low‐dose group was 1.28 (95% confidence interval [95% CI]: 0.94–1.75). In addition, the risk of hypertension in the high‐dose group (8.1%, 17 of 211) was higher than that in the low‐dose group (1.4%, three of 213), with a risk ratio of 5.72 (95% CI: 1.70–19.23) Interpretation Azithromycin combined with low‐dose methylprednisolone demonstrates non‐inferior efficacy in reducing pulmonary lesions at 6‐month follow‐up compared to combined with high‐dose methylprednisolone while exhibiting a more favorable safety profile.