Construction and validation of an acute exacerbation risk score for fibrotic idiopathic interstitial pneumonia

Introduction

Acute exacerbation (AE) of fibrotic idiopathic interstitial pneumonia (f-IIP) is defined as acute worsening of dyspnoea within 1 month, accompanied by new bilateral ground-glass opacities and/or consolidations superimposed on the background of fibrosis, after the exclusion of alternative explanations such as cardiac failure or fluid overload (1,2). AE represents the most devastating complication during the course of f-IIP. The annual incidence of AE in patients with idiopathic pulmonary fibrosis (IPF) is approximately 9% [95% confidence interval (CI): 7–11%], with cumulative incidences reaching 13% and 19% at 2 and 3 years, respectively (3). AE occurs not only in IPF but also commonly in non-IPF fibrotic IIPs (such as fibrotic nonspecific interstitial pneumonia and unclassifiable IIP) as well as secondary interstitial lung diseases (4,5). Regardless of the underlying IIP type, AE significantly increases short-term mortality risk, with in-hospital mortality exceeding 50% among hospitalised patients and a median survival of only 3–4 months (3,5). Even in survivors, patients often face irreversible pulmonary function decline due to progression of pulmonary fibrosis caused by AE (5).

The pathogenesis of AE remains unclear, and its unpredictable nature poses substantial challenges to clinical management of f-IIP. Identified risk factors for AE include advanced age, impaired baseline pulmonary function [low forced vital capacity (FVC) and diffusing capacity of the lung for carbon monoxide (DLCO)], reduced exercise capacity [shortened 6-min walk distance (6MWD)], decreased resting or lowest oxygen saturation, radiological honeycombing, elevated serum lactate dehydrogenase (LDH), and increased neutrophil-to-lymphocyte ratio (NLR) (3,4,6,7). However, the predictive ability of single factors varies significantly across different studies and IIP subtypes, and currently no validated, standardised prediction tool for AE is available. Risk assessment in clinical practice relies largely on physicians’ subjective impressions, lacking reproducibility and comparability, which severely restricts early identification of high-risk populations and the formulation of intervention strategies (8).

In contrast, composite scoring systems integrating multiple clinical variables have demonstrated important value in prognostic evaluation of IIPs. The Gender-Age-Physiology (GAP) (incorporating FVC and DLCO) shows good discrimination for survival prognosis in IPF (9-11); the composite physiology index (CPI) improves the accuracy of survival prediction by correcting for the effect of emphysema (12). However, these models were designed specifically for overall survival (OS) prediction and are not applicable to AE prediction. Recent studies have begun to explore AE-specific prediction tools: the Korean national IPF cohort (KICO) constructed an AE prediction model based on FVC, 6MWD and home oxygen therapy use, with an area under the curve (AUC) of 0.746 (13); Japanese researchers developed a decision tree model integrating baseline FVC and oxygenation index [partial pressure of arterial oxygen (PaO₂)/fraction of inspired oxygen (FiO₂)] at AE onset, with a C-index of 0.735 for predicting 90-day mortality (14). Nevertheless, existing models have notable limitations: most target only IPF populations, neglecting patients with non-IPF f-IIPs; they rely on single-dimension data (pulmonary function alone or laboratory indicators alone), lacking multidimensional integration of clinical, radiological and laboratory parameters; and they have not been adequately validated for applicability and calibration across different IIP subtypes.

The heterogeneity of f-IIP necessitates that AE risk prediction balances disease specificity with clinical generalisability. Current evidence indicates that AE risk stratification should shift from single-parameter assessment to comprehensive models integrating multidimensional data, combining demographic characteristics, physiological indicators, laboratory biomarkers and radiological findings (13). This comprehensive approach may facilitate early identification of high-risk patients, closer clinical monitoring, timely reassessment of symptoms and oxygenation status, and prompt evaluation when AE is suspected. Furthermore, extending AE risk tools to survival prognostic stratification can provide more comprehensive information for clinical decision-making, achieving the clinical application value of “one tool, dual functions”.

Therefore, this study aims to develop and internally validate a novel, clinically accessible risk scoring system for AE of f-IIP. We hypothesise that a multidimensional model integrating clinical indicators, laboratory biomarkers and radiological parameters will outperform existing single-dimension prediction methods. Specific objectives include: (I) identification of independent risk factors for AE through multivariate survival analysis; (II) construction of an intuitive nomogram-based prediction model and transformation into a simplified integer score; (III) evaluation of model discrimination, calibration and clinical utility through Bootstrap internal validation; and (IV) validation of the dual predictive value of this score for both AE incidence and OS, with particular attention to performance differences between IPF and non-IPF fibrotic IIP subgroups. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0623/rc).

Methods

Study design

This single-centre, retrospective, cohort study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study protocol was approved by the Ethics Committee of Beijing Friendship Hospital, Capital Medical University. The requirement for informed consent was waived because of the retrospective observational design. We retrospectively analysed the medical records of 638 patients with f-IIP who were diagnosed and treated at our institution between 2021 and 2025.

Patients enrolled

Patients inclusion criteria were: (I) age ≥18 years; (II) diagnosis of idiopathic interstitial pneumonia (IIP) in accordance with American Thoracic Society/European Respiratory Society guidelines (15,16); (III) definite fibrotic changes on baseline high-resolution computed tomography (HRCT): reticulation, honeycombing, traction bronchiectasis or traction bronchiolectasis; and (IV) complete follow-up records. Exclusion criteria were: (I) presence of identifiable secondary or non-idiopathic causes of interstitial lung disease, including connective tissue disease-associated interstitial lung disease, interstitial pneumonia with autoimmune features, hypersensitivity pneumonitis, sarcoidosis, occupational or environmental exposure-related interstitial lung disease, and drug-induced interstitial lung disease; (II) concomitant other severe pulmonary diseases, such as malignancy or chronic obstructive pulmonary disease, severe cardiovascular disease, or malignancy; (III) recent severe acute infection, uncontrolled tuberculosis, or pulmonary haemorrhage; and (IV) incomplete patient data, lack of necessary follow-up data, or inability to obtain biomarker or imaging data.

Group definitions: patients were classified into IPF and other f-IIP groups according to IIP subtype. The IPF group included patients whose diagnosis of IPF was confirmed by clinical records and/or multidisciplinary discussion (MDD), after exclusion of definite secondary causes, and who fulfilled any of the following criteria: (I) HRCT showing a usual interstitial pneumonia (UIP) pattern; (II) HRCT showing a probable UIP pattern with a final clinical/MDD diagnosis of IPF; or (III) atypical HRCT findings with surgical lung biopsy/cryobiopsy showing UIP and an MDD conclusion of IPF. Patients who met the diagnostic criteria for f-IIP but did not fulfil the diagnostic criteria for IPF were classified into the other f-IIP group. This group mainly included patients with fibrotic nonspecific interstitial pneumonia and unclassifiable f-IIP, as determined based on clinical records, HRCT findings, and, when available, pathological findings through MDD.

Definition of AE: AE was defined according to previously reported criteria for AE of IIP (16,17). Specifically, AE was defined as acute worsening or development of dyspnoea within 1 month, accompanied by newly developed bilateral ground-glass opacities and/or consolidations superimposed on pre-existing fibrotic changes on HRCT, after exclusion of alternative explanations such as cardiac failure or fluid overload.

Univariate and multivariate Cox proportional hazards analysis of risk factors for AE in overall f-IIP patients

The first occurrence of AE during follow-up was defined as the outcome event. Follow-up started at the first visit and ended at the date of first AE occurrence or the last follow-up. Patients without AE were censored at the last follow-up. Baseline variables were obtained from the first-visit records and included demographic and clinical characteristics, symptoms, pulmonary function indices, 6MWD, 6-minute walk test (6MWT Min SpO₂), HRCT-derived imaging features, laboratory parameters, and treatment status. Treatment status referred to medication use recorded at the first visit. Patients with AE at the first visit were excluded from the risk-factor analysis. Univariate Cox proportional hazards models were then constructed for candidate variables including age, smoking status, pulmonary function, 6MWD, 6MWT Min SpO₂, LDH, albumin, NLR (absolute neutrophil count divided by absolute lymphocyte count), lymphocyte-to-monocyte ratio (LMR) (absolute lymphocyte count divided by absolute monocyte count), and fibrosis score, to calculate hazard ratios (HRs) and 95% CIs. Variables with P<0.10 in univariate analysis were selected as candidate covariates and further entered into multivariate Cox models to estimate the independent effect of each factor on AE risk. Model results were reported as HRs with 95% CIs and P values, and the proportional hazards assumption was checked (16).

Construction of nomogram

A nomogram was constructed based on variables obtained from multivariate Cox analysis. Using the rms package, a Cox regression model was established with the finally included statistically significant variables from multivariate Cox regression, with 1-, 2-, and 3-year AE-free survival probabilities as prediction endpoints. After fitting the model with the cph function and constructing the design matrix, the nomogram function was used to map regression coefficients to point scales, and the nomogram was plotted. In the nomogram, each predictor corresponded to a point bar; points for each patient on each bar were summed to obtain a total score, which was then mapped to 1-, 2-, and 3-year AE-free probabilities for visualisation of individualised risk prediction.

Score construction and internal validation

In the overall f-IIP cohort, a Fine-Gray competing risk model was first used with AE as the outcome and death as a competing event to fit a multivariable model and obtain subdistribution HRs for each predictor. Following previous literature (16), HRs were converted to regression coefficients (log HR), then linearly scaled according to coefficient magnitude and rounded to integer points to construct a simplified risk scoring system. A total risk score was calculated for each patient, and the cohort was divided into low-, intermediate-, and high-risk strata using predetermined thresholds (e.g., 0, 1, ≥2 points). In the overall cohort, Kaplan-Meier curves for AE-free survival and OS by risk stratum were plotted, and differences between risk strata were compared using log-rank tests. Time-dependent receiver operating characteristic (ROC) analysis was performed at 24 and 36 months to evaluate the discrimination of the risk score for AE, and AUC values were calculated. Using the rms package, bootstrap resampling (1,000 repetitions) based on Cox models was performed, and calibration curves at 2 and 3 years were plotted to assess agreement between predicted probabilities and observed risks. Decision curve analysis (DCA) was used to compare net benefit of the risk score at different threshold probabilities to evaluate its clinical utility.

Subgroup analysis of IPF and other f-IIP groups

Risk scores were further calculated separately in the IPF and other f-IIP subgroups, and their performance was evaluated independently. Time-dependent ROC analysis was used to calculate AUC at 24 months in each subgroup, ROC curves were plotted, and discrimination was compared. Meanwhile, Kaplan-Meier curves for AE occurrence by risk stratum were plotted within each subgroup, and differences in AE risk between risk strata were compared. A “risk score × diagnosis type (IPF vs. other f-IIP)” interaction term was added to the Cox model to test whether a statistical interaction existed between the risk score and clinical subtype, thereby assessing the applicability and stability of the score across different pathological subtypes.

Statistical analysis

Continuous variables were described as mean ± standard deviation, and categorical variables were presented as frequency and percentage. Normality of continuous variables was assessed using the Shapiro-Wilk test. Overall descriptions were provided for the entire f-IIP cohort, and stratified comparisons were performed between the IPF and other f-IIP groups. Continuous variables were compared using independent-samples t-tests, and categorical variables were compared using χ² tests or Fisher’s exact tests, as appropriate. Univariate Cox proportional hazards regression models were used to analyse associations between baseline indicators and AE occurrence risk, and variables with P<0.10 in univariate analysis were further entered into multivariate Cox models. Independent risk factors identified from multivariate analysis were used for subsequent model construction and validation. Bootstrap internal validation with 1,000 resamples was performed. Time-dependent ROC curves and AUC values were used to evaluate model discrimination. Calibration curves were plotted to assess agreement between predicted probabilities and observed outcomes. DCA was used to evaluate clinical net benefit. Kaplan-Meier curves with log-rank tests were used to compare AE-free survival and OS across risk strata. Risk score, disease type, and their interaction term were included in multivariate Cox models to assess whether the predictive effect of the risk score differed between IPF and other f-IIP patients. All statistical analyses were performed using R software, and two-sided P<0.05 was considered statistically significant.

Results

Patient characteristics

A total of 638 patients were enrolled, including 358 with IPF (56.1%) and 280 with other f-IIPs (43.9%). The mean age of the overall population was 61.16±9.35 years, with male predominance (421 men, 66.0%; 217 women, 34.0%). The mean BMI was 24.96±3.93 kg/m². A history of smoking was present in 284 patients (44.5%). The mean disease duration was 3.16±3.15 years. Regarding comorbidities, 156 patients (24.5%) had coronary heart disease, 118 (18.5%) had diabetes mellitus, and 100 (15.7%) had chronic kidney disease.

In terms of clinical manifestations, 210 patients (32.9%) had fever, 480 (75.2%) had cough, and 503 (78.8%) had dyspnoea. Haemoptysis was relatively uncommon in 33 patients (5.2%). Arthralgia was present in 84 patients (13.2%) and rash in 67 patients (10.5%), representing low proportions. Laboratory and imaging findings revealed C-reactive protein (CRP) 7.88±8.46 mg/L, LDH 235.94±59.49 U/L, albumin 37.08±5.52 g/L, white blood cell (WBC) (7.08±2.08)×10⁹/L, and data-driven texture analysis (DTA) fibrosis score 10.82±5.57, among others (Table 1).

Table S1 compared the clinical characteristics between the IPF and other f-IIP groups. No statistically significant differences were found between the two groups in age, sex, BMI, smoking history, disease duration, symptom distribution (fever, cough, dyspnoea, haemoptysis, arthralgia, rash), or most laboratory indicators (CRP, LDH, albumin, white blood cell count, NLR, LMR) (all P>0.05). The prevalence of diabetes mellitus differed significantly between the two groups, with 15.1% in the IPF group and 22.9% in the other f-IIP group (P=0.01). The prevalence of coronary heart disease was higher in the other f-IIP group than in the IPF group (27.9% versus 21.8%), but the difference did not reach statistical significance (P=0.09). Regarding pulmonary function parameters, no statistically significant differences were observed in FVC, DLCO, total lung capacity (TLC), or 6MWT parameters between the two groups (all P>0.05). No statistically significant differences were found in treatment status between the two groups (Table S1).

Incidence of AE

Kaplan-Meier curves revealed significant differences in AE-free survival curves between the IPF and other f-IIP groups (P=0.045), with significantly higher AE occurrence risk in the IPF group than in the other f-IIP group (Figure 1). At 60 months of follow-up, the AE-free probability was approximately 0.20 in the IPF group and approximately 0.4 in the other f-IIP group.

Figure 1 AE-free survival in IPF and other f-IIP patients. Kaplan-Meier curves showing the probability of AE-free survival over 60 months in patients with IPF (red) and other f-IIP (blue). AE, acute exacerbation; f-IIP, fibrotic idiopathic interstitial pneumonia; IPF, idiopathic pulmonary fibrosis.

Risk factors for AE in f-IIP patients

Univariate Cox proportional hazards models indicated that multiple baseline indicators were significantly associated with AE occurrence risk. Each 1-year increase in age was associated with elevated AE risk (HR =1.022, 95% CI: 1.009–1.035, P=0.001). Patients with a smoking history revealed higher AE risk (HR =1.264, 95% CI: 1.008–1.587, P=0.043). Compared with IPF, other f-IIP patients demonstrated lower AE risk (HR =0.790, 95% CI: 0.627–0.995, P=0.046). Pulmonary function and exercise capacity-related indicators revealed negative associations with AE risk: FVC %pred (HR =0.972, 95% CI: 0.963–0.980, P<0.001), DLCO %pred (HR =0.983, 95% CI: 0.974–0.991, P<0.001), total lung capacity percent predicted (TLC %pred) (HR =0.991, 95% CI: 0.984–0.999, P=0.02), 6MWT distance (HR =0.999, 95% CI: 0.998–1.000, P=0.006), and 6MWT lowest SpO₂ (HR =0.961, 95% CI: 0.936–0.986, P=0.002), all indicating that higher levels were associated with lower AE risk. Regarding laboratory and imaging findings, albumin was identified as a protective factor (HR =0.976, 95% CI: 0.956–0.996, P=0.01), whereas NLR (HR =1.049, 95% CI: 1.024–1.074, P<0.001), LMR (HR =1.077, 95% CI: 1.031–1.126, P<0.001), and DTA fibrosis score (HR =1.056, 95% CI: 1.035–1.078, P<0.001) were associated with increased AE risk. Among treatment factors, corticosteroid therapy was associated with increased AE risk (HR =1.536, 95% CI: 1.223–1.929, P<0.001), and antifibrotic therapy revealed a borderline protective trend (HR =0.805, 95% CI: 0.640–1.014, P=0.06) (Table 2).

After adjustment for relevant covariates in multivariate Cox models, the following factors remained independently associated with AE occurrence: age (HR =1.022, 95% CI: 1.009–1.036, P=0.001), DTA fibrosis score (HR =1.049, 95% CI: 1.015–1.083, P=0.004), and NLR (HR =1.002, 95% CI: 1.001–1.048, P=0.042) were independently associated with increased AE risk; whereas higher DLCO %pred (HR =0.983, 95% CI: 0.975–0.991, P<0.001), higher FVC %pred (HR =0.970, 95% CI: 0.962–0.979, P<0.001), and higher 6MWT Min SpO₂ (HR =0.964, P=0.03) were independently associated with decreased AE risk (Table 3).

Construction of AE prediction nomogram and risk score

A nomogram prediction model for AE in f-IIP patients was constructed based on independent risk factors identified through multivariate Cox analysis (Figure 2A). This nomogram incorporated six variables: the model included six independent predictive variables: age, DLCO %pred, FVC %pred, 6MWT minimum SpO₂, DTA fibrosis score, and NLR. The results revealed that AE-free probabilities in all three groups decreased gradually over time, but the decrease was most pronounced in the high-risk group (blue), whose curve separated from the low/intermediate-risk groups from early follow-up and remained at the lowest level throughout the entire follow-up period; AE-free curves for the low-risk group (red) and intermediate-risk group (green) were generally similar and significantly higher than those for the high-risk group. Log-rank tests indicated statistically significant differences among the three curve, demonstrating that this risk score effectively discriminated populations with different AE occurrence risks (P<0.001, Figure 2B). Patients with higher risk scores revealed lower OS, with significant differences among the three survival curves (P=0.001) (Figure 2C).

Figure 2 AE prediction nomogram and risk score construction. (A) Nomogram constructed based on variables obtained from multivariate Cox analysis. (B) Kaplan-Meier curves showing AE occurrence curves by risk stratum. (C) Kaplan-Meier curves showing overall survival by risk stratum. 6MWT Min SpO₂, minimum oxygen saturation during the 6-minute walk test; AE, acute exacerbation; DLCO %pred, diffusing capacity of the lung for carbon monoxide percent predicted; DTA, data-driven texture analysis; FVC %pred, forced vital capacity percent predicted; NLR, neutrophil-to-lymphocyte ratio.

Next, we performed validation of the risk score. Time-dependent ROC curves at 24 months revealed that the AUC for AE prediction was 0.843, indicating good discrimination ability of the model for AE occurrence within 24 months (Figure 3A). In the 24-month calibration curve, the predicted AE probabilities from the model were generally close to the ideal 45° reference line compared with the actually observed AE occurrence probabilities, suggesting good agreement between predicted and true risks; mild deviation existed in some high predicted probability intervals, but overall calibration performance remained stable (Figure 3B). Decision curves revealed that across a wide range of threshold probabilities, the net benefit of the Cox model (red line) was higher than both the “treat all” and “treat none” strategies, suggesting that risk stratification and intervention decisions based on this model could yield greater clinical net benefit and had potential clinical application value (Figure 3C).

Figure 3 Risk score validation. (A) Time-dependent ROC curve at 24 months. (B) Calibration curve at 24 months. (C) Decision curve. AE, acute exacerbation; AUC, area under the curve; ROC, receiver operating characteristic.

Subgroup analysis of IPF and other f-IIP

To further validate the applicability of the risk scoring system across different f-IIP subtypes, we performed subgroup analyses for IPF and other f-IIP, as shown in Figure 4. In the IPF subgroup, the risk score model demonstrated moderate predictive performance. Figure 4A revealed that the AUC of the time-dependent ROC curve at 24 months was 0.729, suggesting moderate discrimination ability of the model for AE occurrence risk in IPF patients. The calibration curve in Figure 4B revealed acceptable agreement between predicted and observed probabilities in the low-risk interval, but obvious deviation in the intermediate-high risk intervals, presenting a “J-shaped” or “S-shaped” curve, indicating suboptimal calibration of the model in IPF patients, with potential overestimation or underestimation of actual AE occurrence probability especially when predicting intermediate risk. In the other f-IIP subgroup, the risk score model revealed superior predictive performance. Figure 4C revealed that the AUC of the time-dependent ROC curve at 24 months reached 0.849, significantly higher than that in the IPF subgroup, indicating stronger discrimination ability of the model for AE occurrence risk in other f-IIP patients. The calibration curve in Figure 4D revealed good agreement between predicted and observed probabilities, with the calibration curve approaching the ideal diagonal line, suggesting good calibration of the model in other f-IIP patients. Risk stratification effectively discriminated AE occurrence risk in both subgroups. Figure 4E revealed that in IPF patients, significant differences existed in AE-free survival curves among the low-risk group [green, Risk_grouP = Low(0)], intermediate-risk group [red, Risk_grouP = Intermediate(1)], and high-risk group [blue, Risk_grouP = High(≥2)] (P=0.01). Figure 4F revealed that in other f-IIP patients, AE-free survival curves among the three risk strata were also significantly different (P=0.01), with risk stratification trends consistent with the IPF group and the high-risk group (blue) showing the worst prognosis. In summary, this risk scoring system demonstrated good discrimination ability and risk stratification efficacy in both IPF and other f-IIP subgroups, with particularly superior performance in other f-IIP patients (AUC: 0.849 versus 0.729). However, the calibration of this model in IPF patients needs improvement, and careful interpretation of predicted probabilities in the intermediate-high risk intervals is warranted in clinical application.

Figure 4 Subgroup analysis of IPF and other f-IIP. (A) Evaluation of risk score performance in the IPF group: ROC curve. (B) Evaluation of risk score performance in the IPF group: calibration curve. (C) Evaluation of risk score performance in the other f-IIP group: ROC curve. (D) Evaluation of risk score performance in the other f-IIP group: calibration curve. (E) Kaplan-Meier curves showing AE occurrence curves by risk stratum in the IPF group. (F) Kaplan-Meier curves showing AE occurrence curves by risk stratum in the other f-IIP group. AE, acute exacerbation; AUC, area under the curve; f-IIP, fibrotic idiopathic interstitial pneumonia; IPF, idiopathic pulmonary fibrosis; ROC, receiver operating characteristic.

To further explore whether an interaction existed between the risk score and disease type (IPF versus other f-IIP), we performed interaction testing in a multivariate Cox model. Table 4 presented the analysis results for the risk score, disease type, and their interaction term. The risk score as a continuous variable was significantly associated with AE occurrence risk (HR =1.066, P<0.001), suggesting that each 1-point increase in the risk score was associated with approximately 6.6% increase in AE occurrence risk. Disease type (IPF versus other f-IIP) revealed no statistically significant effect on AE occurrence risk (HR =1.604, P=0.06). Crucially, the interaction between the risk score and IPF revealed no statistical significance (HR =0.983, P=0.32), indicating that the predictive effect of the risk score on AE occurrence risk did not differ significantly between IPF and other f-IIP patients, that is, the predictive performance of this risk scoring system demonstrated consistency and stability across different disease subtypes. Combined with the subgroup analysis results in Figure 4, although the discrimination of the risk score was superior in other f-IIP patients (AUC =0.849) compared with IPF patients (AUC =0.729), suggesting acceptable discrimination but suboptimal calibration in the IPF population, this subtype might require recalibration or a specific model.

Discussion

This study successfully developed and internally validated a novel risk scoring system for AE based on a retrospective cohort of 638 patients with f-IIP. This scoring system integrated six dimensions: age, pulmonary function indices (FVC %pred, DLCO % pred), 6MWT Min SpO₂, DTA Fibrosis score and NLR, demonstrating good discrimination (AUC =0.843) and calibration for predicting 24-month AE occurrence risk. The results firstly confirmed that IPF patients revealed significantly higher AE occurrence risk than non-IPF fibrotic IIP patients, a finding consistent with previous literature reports (18). Kim et al. revealed that in patients with progressive pulmonary fibrosis (PPF), the AE incidence in the IPF subgroup was approximately twice that in the non-IPF ILD group (8.38 versus 3.21 per 100 patient-years) (4). In this study, Kaplan-Meier curves revealed that at 60 months of follow-up, the AE-free probability was approximately 0.20 in the IPF group and approximately 0.40 in the other f-IIP group, further validating the clinical significance of this difference.

In multivariate analysis, we identified six independent predictive factors. Pulmonary function impairment (decreased FVC and DLCO) was significantly associated with increased AE risk, consistent with the findings of Lee et al. in the KICO (13). Notably, the KICO study constructed an AE prediction model based on FVC, 6MWD, and home oxygen therapy use, with an AUC of 0.746 (13). In comparison, this study incorporated radiological fibrosis score and inflammatory marker NLR as additional predictive dimensions, improving model discrimination to 0.843. Quantitative radiological assessment (DTA score) reflects pulmonary fibrosis burden, whereas NLR serves as a simple indicator of systemic inflammation and may capture the immune imbalance state preceding AE occurrence.

Existing IPF prognostic assessment tools such as the GAP index and CPI were designed primarily for OS prediction and are not applicable to AE-specific prediction. Although these models integrate gender, age, and physiological indices, they lack dynamic assessment of radiological progression and acute inflammatory status (19). In recent years, Japanese researchers developed a decision tree model integrating baseline FVC and oxygenation index (PaO₂/FiO₂) at AE onset, with a C-index of 0.735 for predicting 90-day mortality; however, this model targeted only short-term prognosis after AE occurrence rather than AE occurrence risk prediction (14). The innovations of this study include: (I) applicability to both IPF and non-IPF fibrotic IIP patients, filling the gap of existing models that target only IPF populations; (II) Bootstrap internal validation with 1,000 resamples to ensure model stability and reliability; and (III) DCA confirming positive net benefit across the range of threshold probabilities, suggesting potential value in supporting risk-based monitoring and clinical reassessment.

Subgroup analyses revealed performance differences of this risk scoring system across different IIP subtypes. In other f-IIP patients, the model revealed superior predictive performance (AUC =0.849) with calibration curves approaching the ideal diagonal line; whereas in the IPF subgroup, although discrimination was acceptable (AUC =0.729), calibration was suboptimal with obvious deviation in the intermediate-high risk intervals. This finding suggests that IPF patients may have more complex AE occurrence mechanisms, and reliance on baseline clinical indicators alone may not fully capture their risk changes. Interaction testing revealed no significant interaction between the risk score and IPF (P=0.32), indicating consistent predictive effects of the scoring system across different disease subtypes, but the IPF subgroup may require additional dynamic monitoring indicators or a specific prediction model.

Notably, the AE prognosis of non-IPF fibrotic IIPs (such as fibrotic NSIP and unclassifiable IIP) has received increasing attention in recent years (20). Kim et al. reported that although the AE incidence in non-IPF ILD was approximately half that in IPF, once AE occurred, the prognosis was equally poor (4). In this study, the other f-IIP group accounted for 43.9% (280/638) of the total cohort, and the risk score performed better in this subgroup, suggesting that this tool has important risk stratification value for this clinically heterogeneous population. The nomogram and risk scoring system constructed in this study have clinical application potential for “one tool, dual functions”. On one hand, this score can help predict AE occurrence risk and identify high-risk patients who may require closer follow-up, repeated assessment of respiratory symptoms and oxygenation status, and prompt evaluation when AE is suspected; on the other hand, the score was closely associated with OS (P=0.001) and may provide additional information for long-term prognostic assessment. In the current context where antifibrotic therapy is an established component of disease management for eligible patients with IPF and progressive fibrotic ILD, the proposed risk score is intended to complement, rather than replace, guideline-based treatment decisions. Its primary clinical value lies in helping clinicians identify patients who may require closer monitoring, repeated assessment of respiratory symptoms and oxygenation status, timely evaluation when AE is suspected, and individualized prognostic discussions (21). Moreover, identifying patients at high risk of AE may also provide an opportunity to initiate advance care planning discussions in a timely manner. For patients with a high predicted risk, clinicians may consider discussing goals of care, preferences regarding invasive mechanical ventilation, and the potential role of palliative care, given the poor prognosis associated with intensive care unit (ICU) admission and mechanical ventilation in patients with ILD or IPF (22,23). This perspective is clinically relevant because patient-centred discussions regarding treatment expectations, ventilatory support, transplant candidacy, and palliative care are important for aligning subsequent management with patients’ values and prognosis (23). From a clinical practice perspective, the six indicators required by this scoring system are routinely accessible parameters: Age, DLCO %pred, FVC %pred, 6MWT Min SpO₂, DTA fibrosis score and NLR. Among them, the DTA score is based on quantitative HRCT analysis; Although its prognostic value was supported, its accessibility in primary healthcare institutions may be limited. Future development of a simplified scoring version or exploration of artificial intelligence-based automated imaging analysis tools may be considered to improve clinical applicability.

This study has the following limitations. Firstly, as a single-centre retrospective study, it carries inherent risks of selection bias and information bias, and the generalisability of results requires multicentre external validation. Secondly, the study period spanned different phases of antifibrotic drug clinical application; Although treatment status revealed no statistically significant difference between groups, the effect of drug use on AE risk may have varied over time. Thirdly, despite bootstrap internal validation showing good model stability, the lack of an independent external validation cohort remains an important step in predictive model development. Additionally, the calibration issue in the IPF subgroup suggests that for this specific population, integration of dynamic pulmonary function changes, serum biomarker trajectories, or machine learning algorithms may be needed to improve prediction accuracy.

Conclusions

This study successfully developed and internally validated a multidimensional clinical index-based risk scoring system for AE of f-IIP. This score integrates age, pulmonary function (FVC, DLCO), exercise capacity (6MWT Min SpO₂), radiological fibrosis burden (DTA), and inflammatory status (NLR), and demonstrates good discrimination (AUC =0.843) and calibration in the overall population, with effective prediction of OS. Although calibration in the IPF subgroup needs improvement, performance in other f-IIP patients was excellent. This scoring system provides a simple, accessible tool for AE risk stratification and clinical decision-making in f-IIP patients, with potential clinical application value. Future prospective multicentre studies are needed for external validation, and exploration of optimised models integrating novel biomarkers and dynamic monitoring data is warranted.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0623/rc

Data Sharing Statement: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0623/dss

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0623/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0623/coif). The 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. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Beijing Friendship Hospital, Capital Medical University. The requirement for informed consent was waived because of the retrospective observational design.

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. Raghu G, Remy-Jardin M, Richeldi L, et al. Idiopathic Pulmonary Fibrosis (an Update) and Progressive Pulmonary Fibrosis in Adults: An Official ATS/ERS/JRS/ALAT Clinical Practice Guideline. Am J Respir Crit Care Med 2022;205:e18-47. [Crossref] [PubMed]
2. Yoo JW, Kim J, Song JW. Impact of the revised definition on incidence and outcomes of acute exacerbation of idiopathic pulmonary fibrosis. Sci Rep 2022;12:8817. [Crossref] [PubMed]
3. Wang Y, Ji Z, Xu B, et al. The incidence of acute exacerbation of idiopathic pulmonary fibrosis: a systematic review and meta-analysis. Sci Rep 2024;14:21080. [Crossref] [PubMed]
4. Kim MJ, Yang J, Song JW. Acute exacerbation of progressive pulmonary fibrosis: incidence and outcomes. Respir Res 2024;25:415. [Crossref] [PubMed]
5. Zhang Y, Cen Z, Ding Q, et al. Prognostic significance of acute exacerbations and usual interstitial pneumonia in fibrotic interstitial lung disease. Sci Rep 2025;15:21580. [Crossref] [PubMed]
6. Achaiah A, Rathnapala A, Pereira A, et al. Neutrophil lymphocyte ratio as an indicator for disease progression in Idiopathic Pulmonary Fibrosis. BMJ Open Respir Res 2022;9:e001202. [Crossref] [PubMed]
7. Chen Y, Cai J, Zhang M, et al. Prognostic Role of NLR, PLR and MHR in Patients With Idiopathic Pulmonary Fibrosis. Front Immunol 2022;13:882217. [Crossref] [PubMed]
8. Kreuter M, Belloli EA, Bendstrup E, et al. Acute exacerbations in patients with progressive pulmonary fibrosis. ERJ Open Res 2024;10:00403-2024. [Crossref] [PubMed]
9. Suzuki Y, Mori K, Aono Y, et al. Combined assessment of the GAP index and body mass index at antifibrotic therapy initiation for prognosis of idiopathic pulmonary fibrosis. Sci Rep 2021;11:18579. [Crossref] [PubMed]
10. Chandel A, King CS, Ignacio RV, et al. External validation and longitudinal application of the DO-GAP index to individualise survival prediction in idiopathic pulmonary fibrosis. ERJ Open Res 2023;9:00124-2023. [Crossref] [PubMed]
11. Lacedonia D, De Pace CC, Rea G, et al. Machine Learning and BMI Improve the Prognostic Value of GAP Index in Treated IPF Patients. Bioengineering (Basel) 2023;10:251. [Crossref] [PubMed]
12. Lee SH, Park JS, Kim SY, et al. Comparison of CPI and GAP models in patients with idiopathic pulmonary fibrosis: a nationwide cohort study. Sci Rep 2018;8:4784. [Crossref] [PubMed]
13. Lee JH, Jang JH, Kim SY, et al. Development of a Prediction Model for Acute Exacerbation in Idiopathic Pulmonary Fibrosis: A Study of the Korea IPF Cohort Registry. J Korean Med Sci 2025;40:e212. [Crossref] [PubMed]
14. Suzuki T, Hozumi H, Miyashita K, et al. Prognostic classification in acute exacerbation of idiopathic pulmonary fibrosis: a multicentre retrospective cohort study. Sci Rep 2021;11:9120. [Crossref] [PubMed]
15. Arai T, Akira M, Sugimoto C, et al. Seroradiologic prognostic evaluation of acute exacerbation in patients with idiopathic interstitial pneumonia: a retrospective observational study. J Thorac Dis 2020;12:4132-47. [Crossref] [PubMed]
16. Karayama M, Aoshima Y, Suzuki T, et al. A predictive model for acute exacerbation of idiopathic interstitial pneumonias. Eur Respir J 2023;61:2201634. [Crossref] [PubMed]
17. Arai T, Takimoto T, Takeuchi N, et al. Neutrophil-to-Lymphocyte Ratio in Acute Exacerbation of Idiopathic Pulmonary Fibrosis. J Clin Med 2023;12:7446. [Crossref] [PubMed]
18. Song JW, Hong SB, Lim CM, et al. Acute exacerbation of idiopathic pulmonary fibrosis: incidence, risk factors and outcome. Eur Respir J 2011;37:356-63. [Crossref] [PubMed]
19. Ye H, Gong W, Yuan P, et al. A supported decision-making model for idiopathic pulmonary fibrosis based on feature screening and optimized neural network. Sci Rep 2025;16:207. [Crossref] [PubMed]
20. Tsubouchi K, Hamada N, Tokunaga S, et al. Survival and acute exacerbation for patients with idiopathic pulmonary fibrosis (IPF) or non-IPF idiopathic interstitial pneumonias: 5-year follow-up analysis of a prospective multi-institutional patient registry. BMJ Open Respir Res 2023;10:e001864. [Crossref] [PubMed]
21. Jovanovic D, Mogulkoc N, Sterclova M, et al. Adverse side-effects of antifibrotic drugs and outcome of treatment in idiopathic pulmonary fibrosis. J Thorac Dis 2025;17:10908-23. [Crossref] [PubMed]
22. Huapaya JA, Wilfong EM, Harden CT, et al. Risk factors for mortality and mortality rates in interstitial lung disease patients in the intensive care unit. Eur Respir Rev 2018;27:180061. [Crossref] [PubMed]
23. Mooney JJ, Raimundo K, Chang E, et al. Mechanical ventilation in idiopathic pulmonary fibrosis: a nationwide analysis of ventilator use, outcomes, and resource burden. BMC Pulm Med 2017;17:84. [Crossref] [PubMed]