Meaningful thresholds for change in quantitative computed tomography–derived fibrosis extent in idiopathic pulmonary fibrosis

Abstract

Rationale

Validated thresholds for clinically meaningful change in quantitative computed tomography (QCT) for fibrosis extent across algorithms remain unclear in idiopathic pulmonary fibrosis (IPF).

Objectives

To evaluate minimal clinically important difference (MCID) for 1-year change in QCT metrics in IPF using different algorithms than previously reported and to identify prognostically meaningful thresholds for effective risk stratification.

Methods

This multicenter retrospective study included IPF patients with baseline and 1-year follow-up CT. Minimal clinically important differences for change in fibrosis score (ΔFS) derived from fully automated CT quantification were estimated using an anchor-based approach with 1-year changes in FVC and diffusing capacity for carbon monoxide (DLco), and a prognostic threshold was estimated using maximally selected log-rank statistics. Prognostic significance of ΔFS was evaluated in discovery and validation cohorts.

Results

The discovery cohort included 524 patients (mean age = 66.8 years; 79% men), and the external validation cohort included 224 patients (mean age = 69.6 years; 83% men). Anchor-based MCIDs were 2.72% for FVC and 4.52% for DLco, with a prognostic threshold of 4.05%. In the discovery cohort, 1-year ΔFS were independently associated with transplant-free survival (TFS), with greater risk when ΔFS exceeded MCID or prognostic thresholds (all, P < .001). In the validation cohort, ΔFS remained associated with TFS (adjusted hazard ratio [aHR] = 1.11; 95% CI, 1.04-1.18), and all prespecified ΔFS thresholds were prognostic for overall TFS, with the prognostic threshold of ΔFS ≥ 4.05% showing the most consistent associations with overall (aHR = 2.78; 95% CI, 1.36-5.68) and 3-year TFS (aHR = 2.88; 95% CI, 1.11-7.48).

Conclusion

One-year change in FS and its thresholds were significantly associated with TFS, supporting annual CT follow-up with predefined QCT thresholds for monitoring and risk stratification in IPF.

At a Glance Commentary.

Scientific Knowledge on the Subject: Quantitative computed tomography (QCT) analysis has emerged as an objective measure for idiopathic pulmonary fibrosis (IPF) assessment, offering advantages over traditional visual interpretation; however, validated thresholds for clinically meaningful change and optimal follow-up intervals in QCT for fibrosis extent across algorithms, populations, and follow-up protocols remain unclear, limiting effective clinical implementation and monitoring.

What This Study Adds to the Field: This study defines clinically meaningful thresholds of fibrosis progression measured by QCT in IPF. Across multiple anchor-based and distribution-based approaches, minimal clinically important difference (MCID) estimates for worsening clustered between approximately 2% and 5% (1.70%-4.94%), defining a clinically meaningful range of fibrosis progression on serial CT. Anchor-based MCIDs corresponded to a 2.72% increase in fibrosis score (FS) at 1 year when referenced to absolute 5% FVC decline and a 4.52% increase when referenced to absolute 10% diffusing capacity for carbon monoxide decline. All prespecified dichotomized ΔFS thresholds were prognostic in both discovery and validation cohorts with overall transplant-free survival (TFS), with a prognostic threshold of 4.05% showing the most consistent and reproducible associations with overall and 3-year TFS in the external validation cohort. These findings provide a practical basis for incorporating annual QCT into clinical follow-up to detect clinically important disease progression and to improve risk stratification in patients with IPF.

Introduction

Idiopathic pulmonary fibrosis (IPF), a chronic, progressive interstitial lung disease (ILD) with an unknown cause, is characterized by an irreversible decline in pulmonary function.¹ Once diagnosed, the prognosis is poor, with a median survival of 3-5 years.² While antifibrotic therapies slow lung function decline,^3,4 disease progression varies significantly among patients, making accurate assessment and longitudinal monitoring essential for patient care and therapeutic decision-making.⁵

Progression of IPF is typically marked by worsening fibrotic changes on imaging, physiological decline measured by pulmonary function tests (PFTs), exercise capacity reduction, and worsening respiratory symptoms.⁶ Forced vital capacity (FVC) decline has been the primary endpoint in clinical trials, leading to antifibrotic drug approvals.^7,8 While high resolution computed tomography (HRCT) plays an essential role in IPF diagnosis and monitoring,⁹ its role as a reproducible endpoint has remained limited. Traditional visual analysis of HRCT faces limitations due to interobserver variation and insufficient precision for tracking disease progression over time.¹⁰ To address these limitations, quantitative CT (QCT) analysis tools have emerged as objective measures of disease extent, with several having received regulatory approval for clinical use.^11,12 However, for effective implementation in clinical practice, establishing thresholds for meaningful change is important. The minimal clinically important difference (MCID) defines the smallest change in a measured outcome that would be considered clinically meaningful and potentially influence therapeutic decision-making, and several studies have proposed MCID for FVC and 6-minute walk distance test values.^13-15 Regarding CT quantification, while previous research using machine learning–based CT lung fibrosis quantification has established an MCID of 3.4% based on 1-year FVC decline,¹⁰ another suggested a 4% threshold change at 6 months for predicting disease progression, offering potential advantage for earlier disease stratification.¹⁶ More recently, a study using data-driven texture analysis (DTA) demonstrated that a 5% increase over 1 year was associated with both accelerated FVC decline and worse transplant-free survival (TFS), with these associations validated in an external cohort.¹⁷ However, standardization of QCT-derived metrics remains an unmet need. Reported thresholds vary due to heterogeneity in patient populations, follow-up intervals, and baseline severity. In addition, quantification software differs by algorithm and feature selection, making it challenging to establish universally applicable thresholds. This highlights the need for validation across different algorithms as well as individual validation of each algorithm. Additionally, although annual CT is often performed when clinical worsening is suspected in patients with IPF,¹⁸ there is no validated optimal interval of CT follow-up.

We aimed to estimate MCID thresholds for 1-year change in QCT metrics in IPF using different quantification algorithms than previously reported and to identify prognostically meaningful thresholds for effective monitoring and risk stratification using discovery and independent validation cohort.

Methods

This retrospective study was approved by institutional review boards of the participating centers (IRB number: 2024-0462, 2025-09-040), and the requirement for written informed consent was waived.

Study population

For the discovery cohort, we retrospectively identified patients with IPF at Asan Medical Center, a 2700-bed tertiary referral academic hospital in Seoul, Republic of Korea, with a dedicated ILD clinic, between January 2011 and April 2020 who had baseline PFT and volumetric HRCT. Patients who died before the 1-year follow-up, lacked 1-year follow-up volumetric HRCT and/or PFTs, or had poor CT image quality that precluded accurate quantitative analysis were excluded. For the external validation cohort, we identified patients with IPF from Samsung Medical Center, a tertiary referral center in Seoul, Republic of Korea, diagnosed between July 2008 and December 2022 who underwent baseline PFT and had paired volumetric HRCT examinations at baseline and approximately 1-year follow-up. Patients with missing baseline PFT measures and poor CT image quality were excluded (see Figure 1).

Figure 1.

Flowchart of patient inclusion. (A) Discovery cohort. (B) External validation cohort. Abbreviations: CT, computed tomography; DLco, diffusing capacity for carbon monoxide; HRCT, high-resolution CT; IPF, idiopathic pulmonary fibrosis; PFT, pulmonary function test.

Clinical data collection and survival outcomes

Demographic and clinical data extracted from electronic medical records included age, sex, body mass index, smoking history, baseline and follow-up PFTs, and use of antifibrotic medication, including the timing of initiation. Pulmonary function measurements included FVC and diffusing capacity for carbon monoxide (DLco) percent predicted, performed within 3 months of the corresponding HRCT, and the closest available measurement was used for analysis. TFS was assessed as the primary outcome. Lung transplant event data were retrieved from institutional database, and TFS was calculated from initial CT date until date of lung transplantation (event) or death (event) or last known date of participant being alive (censored).

HRCT analysis

HRCT scans were performed during full inspiration and without intravenous contrast administration, using multidetector CT scanners. Fibrosis extent (fibrosis score [FS]) was quantified using deep learning software (AVIEW, version 1.1.44.27; Coreline Soft, Seoul, Republic of Korea) on baseline and follow-up volumetric HRCT.^19,20 Minimum requirements for QCT included inspiratory, noncontrast acquisition with axial slice thickness and interval ≤1.5 mm, reconstructed with a sharp kernel.¹ All patients in the discovery cohort were scanned using identical slice thickness and intervals from the same vendor (Siemens Healthineers, Erlangen, Germany; 1.0-mm slice thickness and interval). In the validation cohort, protocols were more heterogeneous, with 155 (69.2%) patients scanned using consistent parameters; most were imaged on GE scanners (GE Healthcare, Chicago, IL, USA; 1.25-mm slice thickness and interval). Detailed CT acquisition parameters and quantification methodology are described in the online supplement (Table E1).

Statistical analysis

Patient characteristics were analyzed using standard summary statistics. To assess potential selection bias, baseline characteristics were compared between included patients and those excluded due to missing 1-year follow-up volumetric HRCT. Minimal clinically important difference thresholds for changes in FS (ΔFS) were evaluated for 1-year follow-ups using anchor-based methods relative to established clinically significant changes in FVC and DLco and distribution-based methods¹⁰ in the discovery cohort (Supplementary material). Given that MCIDs reflect the smallest clinically meaningful change rather than optimal prognostic discrimination, we also identified a prognostic threshold for ΔFS using maximally selected rank statistics, with overall TFS as the outcome.²¹

Prognostic significance of both MCID and the prognostic threshold for ΔFS were evaluated using univariable and multivariable Cox proportional hazard analysis models adjusted for age, use of antifibrotic agent, and baseline FVC to evaluate the association with overall TFS and 3-year TFS. Given that median survival in IPF is typically 3-5 years, 3-year TFS was prespecified as a clinically meaningful midterm outcome.^22,23 Use of antifibrotic agent was modeled as a time-dependent covariate to account for the timing of treatment initiation. To evaluate the added prognostic value of ΔFS beyond established clinical predictors, we assessed the improvement in C-index when adding baseline FS and ΔFS to the gender-age-physiology (GAP) stage model.²⁴ Harrell C-indices were calculated.²⁵ Comparisons of C-indices between nested models were performed using a z-test to derive P-values, as previously described.^25,26 Survival estimates were performed using the Kaplan–Meier method stratified by MCID of ΔFS and compared by log-rank test. Interaction terms were used to test whether the prognostic impact of predefined threshold of ΔFS differed by baseline FS, antifibrotic treatment initiation within 1 year of baseline, and CT protocol or vendor differences. All statistical analyses were performed using SAS software version 9.1.3 (SAS Inc., Cary, NC, USA) and R software (version 4.1.0, R Foundation for Statistical Computing, Vienna, Austria). A P-value <.05 was considered statistically significant.

Results

Study patients’ baseline demographic and clinical characteristics

In the discovery cohort, 914 patients with IPF were screened. Patients who died before the 1-year follow-up (n = 30), lacked 1-year volumetric HRCT (n = 347) or PFTs (n = 12), or had poor CT image quality (n = 1) were excluded, leaving 524 patients (mean age, 66.8 ± 7.7 years; n = 414 [79.0%] men) with complete baseline and 1-year follow-up data. Compared with included patients, those excluded because of missing 1-year follow-up volumetric HRCT (n = 347) had lower baseline FVC and DLco and higher baseline FS (all, P < .001) (Table E2). In the external validation cohort, 228 patients with available baseline PFT and volumetric HRCT at both baseline and 1-year follow-up were screened; poor baseline CT image quality (n = 1) or those with missing DLco values (n = 3) were excluded, resulting in 224 patients in the final cohort (Figure 1 and Table 1).

Table 1.

Baseline characteristics and follow-up outcomes of study patients with 1-year follow-up CT.

Variables	Discovery cohort (n  = 524)	Validation cohort (n = 224)	P-value
Age, mean (SD), y	66.8 (7.7)	69.6 (7.5)	<.001
Male sex, No. (%)	414 (79.0)	185 (82.6)	.26
Body mass index, mean (SD), kg/m2	24.7 (3.0)	24.7 (3.4)	.94
History of smoking, No. (%)	357 (68.1)	170 (75.9)	.03
Baseline FVC, predicted %, median (IQR)	74 (62-85)	76 (66-84)	.12
Baseline DLco, predicted %, median (IQR)	59 (47-71)	66 (53-75)	<.001
Baseline GAP stage, No. (%)			.02
GAP stage 1	313 (59.7)	126 (56.2)
GAP stage 2	174 (33.2)	92 (41.1)
GAP stage 3	37 (7.1)	6 (2.7)
Baseline QCT, median (IQR)
Fibrosis score, % of whole lung volume	8.90 (4.86-17.20)	5.88 (3.04-11.72)	<.001
Reticulation, % of whole lung volume	6.69 (3.84-11.71)	4.93 (2.83-9.86)	<.001
Honeycombing, % of whole lung volume	1.27 (0.19-4.36)	0.27 (0.04-1.45)	<.001
Antifibrotics usage, No. (%)	441 (84.2)	178 (79.5)	.12
Antifibrotics usage within 1 year, No. (%)	105 (20.0)	149 (66.5)	.001
Event of lung transplantation, No. (%)	23 (4.4)	0 (0)	.001
Event of lung transplantation within 3 years, No. (%)	6 (1.2)	0 (0)	.10
Event of death, No. (%)	314 (59.9)	36 (16.1)	<.001
Event of death within 3 years, No. (%)	99 (18.9)	19 (8.5)	<.001

Abbreviations: DLco, diffusing capacity for carbon monoxide; FVC, forced vital capacity; GAP, gender-age-physiology; QCT, quantitative computed tomography.

Compared with the discovery cohort, the validation cohort was older (P < .001), had a higher proportion of patients with a history of smoking (P = .03), and had higher baseline DLco (P < .001) (Table 1). Baseline QCT measures for fibrosis were more extensive in the discovery cohort (all, P < .001), whereas antifibrotic therapy initiation within 1 year was more frequent in the validation cohort (66.5% vs 20.0%; P < .001). Notably, in the discovery cohort, patients initiating antifibrotics within 1 year had more advanced baseline disease and greater fibrosis progression (Table E3). Mortality, including 3-year mortality, was higher in the discovery cohort (P < .001). The median time interval between baseline and 1-year follow-up examinations was 13 months (IQR, 12-14) for PFTs and HRCT in both cohorts. The median interval between PFT and HRCT was 0 days (IQR, 0-9) in the discovery cohort and 0 days (IQR, 0-11) in the external validation cohort.

Estimation of MCID for ΔFS at 1-year follow-up CT

One-year ΔFS showed significant inverse correlations with 1-year changes in FVC and DLco in both cohorts, although the correlations were attenuated in the validation cohort (Table E4 and Figure E1). Using these pulmonary function measures as anchors, Table 2 shows the anchor-based MCID estimates for 1-year ΔFS in the discovery cohort. When patients were categorized into groups based on changes in each anchor variable at 1 year, mean ΔFS differed significantly among the groups (for all, P < .001). Intraclass correlation coefficients for ΔFS based on stable (“no change”) patients in each anchor ranged from 0.91 to 0.92 for 1 year, showing excellent reliability. Anchor-based methods yielded MCID estimates of 2.72% when referenced to an absolute 5% decline in FVC and 4.52% when referenced to an absolute 10% decline in DLco.

Table 2.

Anchor-based estimation of MCID of change in FS based on QCT at 1-year follow-ups in discovery cohort.

Anchor	Discovery cohort
	P-value	No. patients	Mean ΔFS at 1-year (95% CI)	Effect size	ICC
FVC absolute change at 1-year	<.001
Δ ≤ −10% (marked decline)		68	7.85 (5.80-9.91)	0.96
−10% < Δ ≤ −5% (slight decline)		102	2.72 (1.83-3.62)	0.30
−5% < Δ ≤ 5% (no change)		292	1.03 (0.56-1.51)	0.11	0.92
5% < Δ (improvement)		62	−1.04 (−3.01 to 0.94)	−0.09
FVC relative change at 1-year	<.001
Δ ≤ −10% (marked decline)		110	6.60 (5.16-8.04)	0.72
−10% < Δ ≤ −5% (slight decline)		89	2.17 (1.32-3.02)	0.24
−5% < Δ ≤ 5% (no change)		226	0.97 (0.46-1.48)	0.10	0.92
5% < Δ (improvement)		99	−0.91 (−2.22 to 0.41)	−0.08
DLco absolute decline at 1-year	<.001
Δ ≤ −15% (marked decline)		74	6.55 (4.62-8.48)	0.71
−15% < Δ ≤ −10% (slight decline)		48	4.52 (2.91-6.14)	0.45
−10% < Δ ≤ 10% (no change)		374	1.09 (0.66-1.52)	0.11	0.91
10% < Δ (improvement)		28	−2.13 (−6.19 to 1.93)	−0.19
DLco relative decline at 1-year	<.001
Δ ≤ −15% (marked decline)		138	5.59 (4.34-6.83)	0.50
−15% < Δ ≤ −10% (slight decline)		50	1.70 (0.23-3.17)	0.18
−10% < Δ ≤ 10% (no change)		263	0.92 (0.51-1.34)	0.11	0.92
10% < Δ (improvement)		73	−0.67 (−2.49 to 1.14)	−0.06

Abbreviations: DLco, diffusing capacity of the lung for carbon monoxide; FS, fibrosis score; FVC, forced vital capacity; ICC, intraclass correlation coefficient; MCID, minimal clinically important difference; QCT, quantitative computed tomography.

For distribution-based methods in the discovery cohort, MCID estimates from distribution-based methods were 4.94% calculated as 0.5 × SD and 2.78% and 2.84% from standard error of measurement approach, respectively (Table 3). Minimal clinically important difference estimates from anchor- and distribution-based methods ranged from 1.70% to 4.94%. In addition to MCID, a prognostic threshold of 4.05% for ΔFS was identified based on risk stratification of overall TFS (Figure E2).

Table 3.

Estimates of the MCID and prognostic threshold for fibrosis score in the discovery cohort.

Threshold type	Estimation methods		Threshold value (n = 524)
MCID	Anchor	FVC absolute change	2.72
MCID	Anchor	FVC relative changea	2.17
MCID	Anchor	DLco absolute change	4.52
MCID	Anchor	DLco relative changea	1.70
MCID	Distribution	0.5 × SD of FS at baseline	4.94
MCID	Distribution	SEM = (SD of FS at baseline) × √(1-ICC)b	2.78
MCID	Distribution	SEM = (SD of FS at baseline) × √(1-ICC)c	2.84
Prognostic threshold	Maximally selected log-rank for TFS		4.05

Abbreviations: DLco, diffusing capacity of the lung for carbon monoxide; FS, fibrosis score; FVC, forced vital capacity; ICC, intraclass correlation coefficient; MCID, minimal clinically important difference; SD, standard deviation; SEM, standard error of measurement; TFS, transplant-free survival.

^aThe mean values of FS in the “slightly decline” group were used as estimates of the MCID for worsening.

^bICC for FVC absolute change (0.920 at 1-year).

^cICC for FVC relative change (0.916 at 1-year).

Association of 1-year change in FS with outcomes

In the discovery cohort, during median 5.71 (IQR, 3.53-7.53) years of follow-up, lung transplantation was performed in 23 (4.4%) patients, including 6 (1.2%) within 3 years. The overall mortality rate was 314 (59.9%), and 99 (18.9%) patients died within 3 years. The 3-year TFS rate was 80.1% (95% CI, 76.7%-83.5%) and median TFS was 6.4 (95% CI, 5.8-6.9) years. In the validation cohort, with a median follow-up of 4.5 (IQR, 3.2-5.4) years, no patients underwent lung transplantation, and a total of 36 (16.1%) patients died, including 19 (8.5%) within 3 years. The 3-year TFS rate was 91.1% (95% CI, 87.2%-95.0%), and the median TFS was not reached.

Univariable results from the discovery cohort are provided in the Supplementary material (Tables E5-E6). In multivariable analyses, baseline FS (adjusted hazard ratio [aHR] = 1.05; 95% CI, 1.04-1.07; P < .001) and 1-year ΔFS (aHR = 1.07; 95% CI, 1.05-1.09; P < .001) were independently associated with worse overall TFS in the discovery cohort. Both associations were reproduced in the validation cohort, with baseline FS (aHR = 1.06; 95% CI, 1.02-1.09; P = .003) and 1-year ΔFS (aHR = 1.11; 95% CI, 1.04-1.18; P < .001) remaining independently prognostic (Table 4). When dichotomized by predefined thresholds, exceeding MCID values based on absolute or relative changes in FVC or DLco was consistently associated with worse overall TFS, with aHRs ranging from 1.67 to 2.60 (all, P < .001) in the discovery and from 2.35 to 2.66 (P = .006-0.01) in the validation cohort. In the validation cohort, ΔFS ≥ 4.05% (prognostic threshold) showed the largest effect estimate (aHR = 2.78; 95% CI, 1.36-5.68; P = .005) with a similar magnitude of effect observed for the DLco-anchored MCID of 4.52% (aHR = 2.66; 95% CI, 1.29-5.49; P = .008). For 3-year TFS, ΔFS ≥ 4.05% demonstrated the most consistent association, remaining statistically significant in the discovery cohort (aHR = 3.70; 95% CI, 2.50-5.48; P < .001) and the validation cohort (aHR = 2.88; 95% CI, 1.11-7.48; P = .03). In contrast, the associations between smaller MCID-based thresholds and survival were attenuated in the validation cohort and showed less consistent statistical significance, despite similar effect directions (Table E7).

Table 4.

Association of lung transplant-free survival and 1-year change in fibrosis score in idiopathic pulmonary fibrosis.

Metrics	Discovery cohort		Validation cohort
	Adjusted HRa (95% CI)	P-value	Adjusted HRa (95% CI)	P-value
Baseline FS	1.05 (1.04-1.07)	<.001	1.06 (1.02-1.09)	.003
ΔFS at 1-yearb	1.07 (1.05-1.09)	<.001	1.11 (1.04-1.18)	<.001
ΔFS ≥ absolute change FVC-anchored MCID (2.72%)	1.87 (1.47-2.37)	<.001	2.60 (1.31-5.16)	.006
ΔFS ≥ relative change FVC-anchored MCID (2.17%)	1.67 (1.33-2.11)	<.001	2.42 (1.23-4.80)	.01
ΔFS ≥ absolute change DLco-anchored MCID (4.52%)	2.60 (2.01-3.36)	<.001	2.66 (1.29-5.49)	.008
ΔFS ≥ relative change DLco-anchored MCID (1.70%)	1.84 (1.47-2.31)	<.001	2.35 (1.19-4.64)	.01
ΔFS ≥ distribution-based MCID (4.94%)	2.46 (1.88-3.21)	<.001	2.36 (1.12-4.95)	.02
ΔFS ≥ prognostic threshold (maximally selected log-rank, 4.05%)	2.52 (1.96-3.24)	<.001	2.78 (1.36-5.68)	.005

Abbreviations: DLco, diffusing capacity of the lungs for carbon monoxide; FS, fibrosis score; FVC, forced vital capacity; HR, hazard ratio; MCID, minimal clinically important difference.

^aAdjusted by age, use of antifibrotic agent (time-dependent covariate), and baseline FVC.

^bΔFS was analyzed as a continuous variable.

In the discovery cohort, the prognostic association of 1-year ΔFS ≥ 4.05% was broadly consistent across baseline FS severity strata (P for interaction = 0.10 for 3-year TFS and 0.06 for overall TFS; Table E8) and by antifibrotic therapy initiation within the first year, although borderline interactions were observed for the absolute change FVC-anchored MCID (2.72%) and prognostic threshold (4.05%) (both P = .05; Table E9). In the external validation cohort, protocol consistency between baseline and follow-up CT did not significantly modify the prognostic value of prespecified 1-year ΔFS thresholds (all P for interaction >.10; Table E10).

Threshold-based risk stratification and incremental prognostic value of FS beyond GAP

In the discovery cohort, Kaplan–Meier analyses consistently demonstrated lower TFS among patients with 1-year ΔFS ≥ 2.72% (FVC-anchored MCID) and among those with 1-year ΔFS ≥ 4.05% (prognostic threshold) (for all comparison, log-rank P < .001; Figure 2). In the validation cohort, both thresholds stratified overall TFS (both P < .001) and 3-year TFS, with more pronounced separation for 3-year TFS using the prognostic threshold (log-rank P = .01 and .003, respectively; Figure 3).

Figure 2.

Kaplan–Meier curves for survival stratified by 1-year change in fibrosis score (FS) in discovery cohort. Kaplan–Meier curves for 3-year transplant-free survival (TFS) (A) and overall TFS (B) stratified by the 1-year change in FS exceeding the FVC-anchored minimal clinically important difference (2.72%) in the discovery cohort. Kaplan–Meier curves for 3-year TFS (C) and overall TFS (D) stratified by the 1-year change in FS exceeding the prognostic threshold (4.05%) in the discovery cohort.

Figure 3.

Kaplan–Meier Curves for survival stratified by 1-year change in fibrosis score (FS) in external validation cohort. Kaplan–Meier curves for 3-year transplant-free survival (TFS) (A) and overall TFS (B) stratified by the 1-year change in FS exceeding the FVC-anchored minimal clinically important difference (MCID) (2.72%) in the external validation cohort. Kaplan–Meier curves for 3-year TFS (C) and overall TFS (D) stratified by the 1-year change in FS exceeding the prognostic threshold (4.05%) in the external validation cohort.

In analyses of the incremental prognostic value of 1-year ΔFS beyond established clinical prediction models for TFS (Table 5), adding baseline FS to GAP stage improved discrimination, increasing the C-index from 0.655 to 0.735 (P < .001), with further gains upon addition of 1-year ΔFS, whether as a continuous variable (C-index, 0.735 to 0.760; P < .001) or dichotomized at the prognostic threshold (C-index, 0.735 to 0.761; P < .001), in the discovery cohort. Figure 4 shows a representative example of 1-year ΔFS exceeding the prognostic threshold. In the validation cohort, adding baseline FS similarly improved discrimination, increasing the C-index from 0.583 to 0.713 (P = .006). Further inclusion of 1-year ΔFS as a continuous variable yielded additional gains, increasing the C-index from 0.713 to 0.764 (P = .04), whereas ΔFS dichotomized at the prognostic threshold produced only a small numerical increase that did not reach statistical significance (C-index, 0.713 to 0.719; P = .77). Results for 3-year TFS were directionally consistent with those for overall TFS in the discovery cohort; however, incremental discrimination gains for ΔFS in the validation cohort did not reach statistical significance (Table E11).

Table 5.

Added prognostic value of change in CT fibrosis score for predicting overall transplantation-free survival outcomes.

Models	Harrell C-index (95% CI)	P-valuea	P-valueb
Discovery cohort (n = 524)
Overall TFS (No. of events = 327)
1) GAP stage	0.655 (0.627-0.682)
2) GAP stage + baseline FS	0.735 (0.706-0.763)	<.001
3) GAP stage + baseline FS + 1-year ΔFS	0.760 (0.732-0.787)		<.001
4) GAP stage + baseline FS + 1-year ΔFS ≥ 4.05	0.761 (0.733-0.788)		<.001
External validation cohort (n = 224)
Overall TFS (No. of events = 36)
1) GAP stage	0.583 (0.490-0.675)
2) GAP stage + baseline FS	0.713 (0.634-0.791)	.006
3) GAP stage + baseline FS + 1-year ΔFS	0.764 (0.693-0.835)		.04
4) GAP stage + baseline FS + 1-year ΔFS ≥ 4.05	0.719 (0.634-0.803)		.77

Abbreviations: CT, computed tomography; FS, fibrosis score; GAP, gender, age, and physiology; MCID, minimal clinically important difference; TFS, transplant-free survival.

^aCompared between GAP stage vs GAP stage + baseline FS model.

^bCompared between GAP stage + baseline FS model vs GAP stage + baseline FS + ΔFS model.

Figure 4.

Progression of fibrosis on high resolution computed tomography (HRCT) at 1-year follow-up in a patient with idiopathic pulmonary fibrosis. (A) A man aged 75 years with a usual interstitial pneumonia pattern on HRCT showed a baseline fibrosis score of 11.66%, comprising 9.34% reticulation and 2.32% honeycombing (rounded values shown in the summary table). (B) At 1-year follow-up, fibrosis had visually progressed (arrows) including traction bronchiolectasis/bronchiectasis with increased subpleural reticulation (colored orange) and honeycombing cysts (colored red). The quantitative fibrosis score was 15.77% (11.39% reticulation and 4.38% honeycombing), corresponding to 4.11% increase compared with baseline CT. Pulmonary function also declined at 1-year follow-up (change in FVC%predicted, –6%; diffusing capacity for carbon monoxide %predicted –13%), and the patient died 2.53 years after baseline.

Discussion

This study estimated clinically meaningful thresholds for 1-year change in QCT-derived FS (ΔFS) in patients with IPF and externally validated their prognostic value. Minimal clinically important difference estimates for worsening clustered between approximately 2% and 5% (1.70%-4.94%) specifically 2.72% anchored to an absolute 5% FVC decline, and 4.52% anchored to an absolute 10% DLco decline. Using overall TFS, we further identified a prognostic threshold of 4.05%. One-year ΔFS, and all prespecified dichotomized ΔFS thresholds were independently associated with overall TFS in discovery and validation cohorts, with a prognostic threshold of 4.05% showing the most consistent and reproducible associations with both overall and 3-year TFS in the external validation cohort. Adding ΔFS to established clinical models (GAP stage and baseline FS) further improved discrimination for overall TFS in both cohorts, although incremental gains were smaller and nonsignificant for the threshold-based model in the validation cohort. Taken together, our results provide evidence supporting the use of annual CT follow-up with predefined QCT thresholds to aid disease monitoring and improve risk stratification in IPF, complementing conventional physiologic indices.

Recent IPF guidelines recommend annual HRCT for the clinical suspicion of disease progression or increased risk of lung cancer;¹⁸ however, evidence defining clinically meaningful change on follow-up CT and its interval remains scarce.^27,28 Although MCID thresholds for quantitative FS on HRCT in IPF have been reported, variabilities in study populations, follow-up intervals, and quantification methods hindered standardization, highlighting the need for further validation. In a clinical trial of 141 IPF patients using QCT, MCID of FS at follow-up HRCT at 54 or 60 weeks was estimated to be 3.4% by Humphries  et al.¹⁰ using FVC as an anchor. Our study estimated the 1-year MCID range for FS as 1.7%-4.9%, depending on the method used for defining MCID. Although slightly lower values were estimated in our study, likely due to shorter follow-up duration or more homogeneous CT acquisition protocols in discovery cohort, these values are comparable to those reported by Humphries  et al. (3.40%-6.44%). Therefore, our findings support previous reports suggesting that an FVC-anchored MCID of approximately 3% in QCT-derived fibrosis extent reflects clinically meaningful progression of functional decline over 12-15 months, irrespective of the quantification software used.

In our current study, the prognostic threshold (ΔFS ≥ 4.05%) provided the most consistent and reproducible associations with survival outcomes. Our findings are consistent with recent work by Koslow and colleagues,¹⁷ who reported that a 1-year increase of 5% in DTA score was associated with lung function decline and worse TFS across independent cohorts. While our study used a different QCT software platform, we identified a similar magnitude of change (4.05%). Direct comparison across tools should be made with caution, as each algorithm (eg, AVIEW [Coreline Soft], DTA, CALIPER, AMFM) is designed to capture distinct radiologic features, and even when labeled as quantifying the same intended feature, the underlying definitions and measurements may differ, leading to variability in reported thresholds.^29,30 Reported thresholds are therefore platform-specific and require their own independent validation prior to clinical implementation; however, the convergence of results across independent methods suggests that a 4%-5% increase in QCT-derived fibrosis extent is prognostically meaningful. Future work applying multiple algorithms to the same datasets will be essential to establish generalizability of MCID and prognostic thresholds.

Traditional visual ILD assessment relies on radiologists’ subjective evaluation,^1,31 which is limited by high inter- and intra-observer variability and was insufficient for accurately evaluating the degree of 3D volume changes.³² These led to the development of automated, computer-based QCT tools for ILD assessment.^33-36 A major hurdle in QCT-based evaluation remains measurement variability. In a prospective study, same-day CT scans with identical parameters showed excellent reproducibility (95% limits of agreement, –0.9% to 1.0%), whereas variability widened substantially with differing reconstruction parameters (–11.3% to 3.9%).³⁷ In this context, the 4.05% threshold identified in our study likely exceeds measurement variability and may therefore reflect true fibrosis progression. However, when CT acquisition parameters are well controlled, even relatively small changes in quantitative fibrosis measures may retain prognostic validity. Together with recent evidence demonstrating high cross-vendor agreement under standardized acquisition conditions (eg, matched peak kilovoltage [kVp] and thin-section volumetric imaging with representative high-frequency lung kernels of each vendor),³⁸ our findings emphasize the necessity of consistent CT acquisition for longitudinal evaluation. When these minimum requirements are met, as was required for inclusion in the present analysis, vendor-related variability may not be the predominant limitation for longitudinal QCT-based risk stratification, whereas larger ΔFS thresholds may be more robust in heterogeneous imaging settings where such standardization cannot be fully ensured. Furthermore, the prognostic significance of ΔFS ≥ 4.05% was consistent across baseline fibrosis severities, with no evidence of effect modification.

Our current study had several limitations. First, the external validation cohort was relatively small, with fewer outcome events and milder baseline disease than the discovery cohort. This likely limited statistical power and reduced the precision and stability of discrimination, particularly for threshold-based models. Second, the retrospective design introduces potential confounding by indication, as antifibrotics were preferentially initiated in patients with more advanced baseline disease. While we modeled treatment as a time-dependent covariate to mitigate bias, our study was not designed for causal inference. Furthermore, borderline interaction results for certain thresholds should be interpreted cautiously due to limited statistical power; however, the prognostic value of ΔFS remained consistent regardless of antifibrotic therapy. Third, excluding patients who died or lacked follow-up CT before 1 year introduced survivor and selection bias. As shown in Table E1, excluded patients had worse baseline lung function and higher fibrosis burden. Consequently, these exclusions may have attenuated the observed associations between early CT changes and mortality, potentially limiting generalizability to those with the most aggressive disease courses. Fourth, although our interaction analysis within the validation cohort did not suggest effect modification by protocol differences among scans meeting the minimum acquisition criteria (thin-section volumetric HRCT reconstructed with sharp kernels), systematic differences in technical parameters including slice thickness and reconstruction interval between cohorts may still contribute to residual heterogeneity not fully captured by this analysis. Finally, our findings are based on measurements obtained with a single QCT platform. Although consistency with prior work using other software supports conceptual generalizability, prospective multiplatform standardization studies are required to confirm transportability across vendors and acquisition protocols. Despite these limitations, our findings suggest that QCT-based prognostic thresholds for change in FS could aid in monitoring IPF and early identification of poor prognosis. This approach could extend to progressive pulmonary fibrosis, potentially transforming ILD trials through objective fibrosis quantification—enabling more precise patient selection likely to benefit from intervention, potentially reducing sample sizes needed for trials, and providing sensitive surrogate endpoints for treatment response.

Conclusion

In conclusion, the MCID for 1-year change in FS was estimated, and a prognostic threshold (4.05%) closely aligned with the DLco-anchored MCID (4.52%) was identified and externally validated. One-year change in FS and its thresholds were significantly associated with TFS, supporting the use of annual CT follow-up with predefined QCT thresholds to improve disease monitoring and risk stratification in IPF.

Author contributions

Conceptualization: H.C.K., J.C. Data curation: D.Y.K., Y.J.O., M.K. Investigation: D.Y.K., Y.J.O., H.Y.L., H.C.K., J.C. Methodology: M.K., H.Y.L., H.C.K., J.C. Project administration: H.Y.L., H.C.K., J.C. Resources: H.J.H., E.J.C., S.M.L, J.H.L, S.P., J.B.S. Supervision: H.Y.L., H.C.K., J.C. Visualization: D.Y.K., M.K. Writing-original draft: D.Y.K., H.Y.L., H.C.K., J.C. Writing-review and editing: D.Y.K., E.J.C., J.H.L, J.B.S., H.Y.L., H.C.K., J.C.

Supplementary material

Supplementary material is available at American Journal of Respiratory and Critical Care Medicine online.

Conflicts of interest

Please see the ICMJE disclosure forms, which have been provided as supplementary material. Activities related to the present article: J.B.S. holds a patent on a method for automatic classifier of lung diseases (Patent No. KR-10-0998630) and has received royalty from Coreline Soft, Co., Ltd. Activities not related to the present article: S.M.L and J.B.S. hold stock/stock options in Coreline Soft, Co., Ltd., Korea. J.C. reports institutional research funding from Coreline Soft, Co., Ltd.

Funding

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. RS-2025-00556764 and No. RS-2025-16067456) and Future Medicine 20*30 Project of the Samsung Medical Center (#SMO1250061).

Data availability

The data of this study are available from the corresponding authors, H.Y.L., H.C.K, and J.C., upon reasonable request.