A Quantitative Imaging Measure of Progressive Pulmonary Fibrosis

Abstract

Rationale

Progressive pulmonary fibrosis (PPF) is common in patients with fibrotic interstitial lung disease (ILD) and leads to high mortality. Although PPF guideline criteria include computed tomography (CT)-based progression, these measures are qualitative and prone to interreader variability. Quantitative computed tomography (qCT) measurements have the potential to overcome this limitation.

Objectives

The objectives of this study were to determine whether changes in qCT measures of pulmonary fibrosis are associated with transplant-free survival (TFS) in a diverse ILD cohort and establish a quantitative computed tomography measure of progressive pulmonary fibrosis (qctPPF).

Methods

A retrospective cohort analysis was performed in individuals with fibrotic ILD, including idiopathic pulmonary fibrosis (n = 350), who underwent serial chest CT for clinical indications. Commercially available software was used to generate qCT measures of pulmonary fibrosis, which were tested for association with 2-year TFS using a multivariable Cox proportional hazards model. Iterative modeling was then performed to develop a composite qctPPF measure. Results were validated in an independent ILD cohort (n = 92).

Measurements and Main Results

Increasing ground-glass opacity and decreasing lung volume showed consistent association with decreased TFS across cohorts when modeled continuously and dichotomously. qctPPF classification was associated with a greater than threefold increased hazard of death or transplant in the test (hazard ratio, 4.41; 95% confidence interval, 2.77–7.03) and validation (hazard ratio, 3.54; 95% confidence interval, 1.62–7.71) cohorts. Agreement between qctPPF and radiologist-determined PPF was poor (κ = 0.20), with qctPPF classification maintaining prognostic significance when discordant with radiologist interpretation.

Conclusions

Changes in qCT measures are associated with clinically relevant outcomes and could improve PPF classification.

At a Glance Commentary

Scientific Knowledge on the Subject

Progressive pulmonary fibrosis (PPF) is common in patients with diverse fibrotic interstitial lung disease (ILD) and leads to high mortality. Current PPF guideline criteria include only qualitative radiologic worsening, but quantitative computed tomography (qCT) measures have the potential to overcome known limitations with interreader variability.

What This Study Adds to the Field

This study shows that changes in qCT measures of pulmonary fibrosis are associated with transplant-free survival in a diverse ILD cohort. We established a quantitative CT measure of PPF, which was associated with a greater than threefold increased hazard of death or transplant in two separate fibrotic ILD cohorts, suggesting that change in qCT measures could improve PPF classification.

Progressive pulmonary fibrosis (PPF) is a devastating and common complication of fibrotic interstitial lung disease (ILD), characterized by declining lung function, worsening respiratory symptoms, and premature death (1–3). Although progression of pulmonary fibrosis characteristically complicates idiopathic pulmonary fibrosis (IPF), it also occurs in other fibrotic ILDs, resulting in morbidity and mortality profiles similar to IPF (1–3). Criteria for classifying PPF were recently proposed, which include a combination of at least two of the following: worsening respiratory symptoms, lung function decline, and radiological progression within 1 year of follow-up despite appropriate management (1).

Inclusion of computed tomography (CT)-based ILD progression in the PPF guideline criteria was supported by studies demonstrating that radiologist-determined baseline extent of fibrosis and increase in extent of fibrosis over time on CT are associated with increased mortality in IPF (4), autoimmune ILDs (5, 6), fibrotic hypersensitivity pneumonitis (fHP) (7), pulmonary sarcoidosis (8), and unclassifiable ILDs (9). Given the need for at least two features of progression when classifying PPF, the well-documented interreader variability among radiologists when interpreting CT fibrotic features (10) may impact PPF classification. Rapidly evolving quantitative computed tomography (qCT) algorithms have the potential to overcome this limitation (10).

Recent studies have shown that automated qCT measures of pulmonary fibrosis are associated with changes in lung function and survival in patients with IPF and other forms of fibrotic ILD (11–17). Although most studies performed to date have used qCT measures from a single time point, several have also demonstrated the prognostic significance of longitudinal change in qCT measures, highlighting their potential to inform PPF classification (13, 18–21). In this study, we aimed to determine whether near-term changes in qCT measures of fibrosis are associated with subsequent 2-year transplant-free survival (TFS) in a diverse ILD cohort. We then used these data to derive a composite quantitative computed tomography–based measure of progressive pulmonary fibrosis (qctPPF) and tested this measure in an independent ILD cohort. Part of this work was presented in abstract form at the 2024 International Colloquium on Airway and Lung Fibrosis in Athens, Greece, and at the 2025 American Thoracic Society International Conference (22).

Methods

Cohorts

A retrospective cohort analysis was performed. Patients diagnosed with fibrotic ILD caused by IPF, connective tissue disease–associated interstitial lung disease (CTD-ILD), fHP, and non-IPF idiopathic interstitial pneumonias (IIPs) evaluated at the University of Michigan (2016–2022) who underwent serial chest CT were eligible for inclusion. Those without follow-up chest CT performed 3–18 months after baseline chest CT were excluded, as were those with missing data for model covariates described below. Findings were tested in an independent cohort from the University of Chicago (2005–2019) with available qCT data. Study-specific protocols were approved at the University of Michigan (HUM00233408) and the University of Chicago (IRB 17-1617 and IRB 14163A).

CT Data Acquisition

Computer-Aided Lung Informatics for Pathology Evaluation and Ratings (CALIPER) software (23) was applied to all chest CT scans meeting the minimum acquisition parameters required for processing (24). CALIPER produces quantitative measures of total lung volume and pulmonary vascular-related structure volume, as well as percent ground-glass opacity, reticular opacity, and honeycombing. Relative change in volume-based measures and absolute change in percent-based measures were calculated using the baseline CT and CT performed closest to 12 months later (allowable range, 3–18 mo).

Survival Analysis

A multivariable Cox proportional hazards regression model was used to test the association between change in quantitative CT measures and subsequent 24-month TFS, defined as the time from the second chest CT to death of any cause, lung transplant, or censoring at 24 months or when lost to follow-up. Potential confounders of the association between change in quantitative CT measures and TFS were collected and included as model covariates, including presence of CTD, sex, race, smoking history, baseline body mass index, age, percent predicted FVC, percent predicted Dl_CO, and CALIPER-derived measures total lung volume, pulmonary vascular-related structure volume, percent ground glass opacity, reticular opacity, and honeycombing.

Changes in qCT measures were modeled continuously and categorically, with the latter established through threshold analysis using the ‘threshold’ package in Stata (StataCorp 2024, Release 18). A single threshold was selected in the test cohort (University of Michigan) on the basis of the Bayesian information criterion and then tested in the validation cohort (University of Chicago). The proportional hazards assumption was checked and satisfied unless otherwise denoted. To corroborate findings, absolute change in restricted mean survival time (RMST) was also estimated using a generalized linear model adjusted for the same covariates. RMST was established by converting TFS to continuous pseudo-observations (25) and normalizing these measures to a range of 0–1, which allows coefficients to be interpreted as absolute change in RMST.

Best-Fit Quantitative CT Measure of PPF

To establish a qctPPF measure, we iteratively fit Cox proportional hazards regression models in the test cohort with combinations of qCT change measures. Model-fitted values were then dichotomized using the threshold analysis describe above. The model providing the highest C-statistic for discriminating subsequent 2-year death or transplant was selected to define qctPPF. We then applied this measure to the validation cohort to assess for ongoing association with 2-year TFS. The Kaplan-Meier estimator was used to plot TFS with a log-rank test to compare survival between qctPPF strata. Cohorts were then combined, and subgroup analysis was performed for common ILD subtypes.

Concordance between qctPPF and Radiologist-determined Progression

The electronic health record for individuals composing the test cohort was manually reviewed to ascertain whether increasing extent of fibrosis was documented on the radiology report. Progression was considered present when one or more of the following terms were mentioned by the interpreting radiologist: new or increased traction bronchiectasis, bronchiolectasis, reticulation, ground-glass opacity with bronchiectasis, honeycombing, and lobar volume loss, according to recently proposed PPF criteria (1). Agreement between radiologist-clinical report determined progression (radPPF) and PPF classification using qctPPF was assessed with a Cohen’s kappa statistic. TFS was plotted according to concordance between the two measures, with a log-rank test used to compare survival between groups.

Progression Analysis

Using serially acquired lung function, 1-year absolute change in FVC and Dl_CO percent predicted from baseline were estimated to determine whether qctPPF correlates with physiological measures of PPF over the same time frame. We then estimated 1-year change in FVC and Dl_CO after qctPPF classification to determine whether this imaging-based measure predicts future change in lung function. Changes in FVC and Dl_CO were estimated using an index pulmonary function test (PFT) and a subsequent PFT performed closest to 12 months later (allowable range, 6–18 mo). Missing FVC because of death or lung transplant was imputed using a value 10% lower than the index value. Sensitivity analyses were performed to investigate the impact of informative missingness on lung function change estimates.

Continuous change in FVC and Dl_CO was compared between qctPPF strata using a two-sample t test. Categorical measures of progression, defined as 1-year absolute decline in FVC percent predicted ⩾5% or Dl_CO percent predicted ⩾10% (1, 26), were compared between qctPPF strata using a chi-square test. TFS was plotted according to concordance between qctPPF and physiological PPF (pftPPF), with a log-rank test used to compare survival between groups.

All statistical analyses were performed using Stata. Continuous variables are presented as means with SDs, and categorical variables are presented as counts with percents. Statistical significance was set at P < 0.05.

Results

Cohorts

Of 816 eligible individuals in the test cohort, 350 met inclusion criteria (Figure 1), with a median time between CT scans of 11.3 months (interquartile range, 7.5–13.1 mo). Ninety-two similarly defined individuals met inclusion criteria in the validation cohort (Figure 1), with median time between CT scans of 11.6 months (interquartile range, 7.6–14.5 mo). The mean age in the test cohort was 66.3 ± 10.7 years, with a slight male predominance (52.6%). The majority were White (83.4%), with a slight predominance of individuals who ever smoked (55.1%). IPF made up the largest ILD subtype at 43.1%, followed by CTD-ILD (33.7%), non-IPF IIP (15.1%), and fHP (8.0%). Mean percent predicted FVC and Dl_CO were 71.9% ± 19.1% and 54.9% ± 19.2%, respectively.

Figure 1.

STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) diagram for the University of Michigan test cohort. CT = computed tomography.

Individuals composing the validation cohort had a mean age of 64.7 ± 10.9 years with a similar breakdown of sex, race, and body mass index compared with the test cohort. Notably, only 19.6% of the validation cohort had CTD-ILD, and more patients had fHP (20.7%) and a history of smoking (62.0%). The mean percent predicted FVC and Dl_CO were 64.2% ± 17.3% and 49.1% ± 19.4%, respectively. Mean baseline and change in qCT measures were similar between cohorts, as was mean change in qCT measures (Table 1). Change in lung volume was moderately correlated with change in ground-glass opacity and reticular opacity, and all other correlations were weak to negligible (see Table E1 in the online supplement).

Table 1.

Baseline Characteristics for University of Michigan Test and University of Chicago Validation Cohorts

	University of Michigan (n = 350)	University of Chicago (n = 92)
Clinical characteristics
Age, yr, mean ± SD	66.3 ± 10.7	64.7 ± 10.9
Male sex, n (%)	184 (52.6)	56 (60.9)
Body mass index (BMI), kg/m2, mean ± SD	30.5 ± 6.5	30.7 ± 7.3
Ever-smoker, n (%)	193 (55.1)	57 (62.0)
Race, n (%)
White	292 (83.4)	73 (79.3)
Black	38 (10.9)	13 (14.1)
Hispanic	5 (1.4)	2 (2.2)
Asian/Pacific Islander	9 (2.6)	3 (3.3)
Other/unknown	6 (1.7)	1 (1.1)
ILD classification, n (%)
IPF	151 (43.1)	40 (43.5)
CTD-ILD	118 (33.7)	18 (19.6)
IIP	53 (15.1)	15 (16.3)
fHP	28 (8.0)	19 (20.7)
PFT characteristics
FVC, L, mean ± SD	2.6 ± 0.9	2.4 ± 0.8
FVC % predicted, mean ± SD	71.9 ± 19.1	64.2 ± 17.3
DlCO, mm/min/mm Hg, mean ± SD	13.0 ± 5.4	10.8 ± 5.8
DlCO % predicted, mean ± SD	54.9 ± 19.2	49.1 ± 19.4
qCT characteristics
Lung volume, L, mean ± SD	3.6 ± 1.2	3.4 ± 1.0
Pulmonary vascular-related structure volume, L, mean ± SD	0.2 ± 0.07	0.2 ± 0.07
% Reticular opacity, mean ± SD	6.4 ± 5.6	6.5 ± 5.2
% Ground-glass opacity, mean ± SD	19.0 ± 20.6	24.2 ± 22.5
% Honeycombing, mean ± SD	0.5 ± 1.3	0.2 ± 0.7

Definition of abbreviations: CTD = connective tissue disease; fHP = fibrotic hypersensitivity pneumonitis; IIP = idiopathic interstitial pneumonia, ILD = interstitial lung disease; IPF = idiopathic pulmonary fibrosis; PFT = pulmonary function test; qCT = quantitative computed tomography.

Survival Analysis

When modeling the association between 2-year TFS and continuous change in qCT measures, each 1% decrease in total lung volume was associated with an increased hazard of 24-month death or lung transplant (hazard ratio [HR], 1.04; 95% confidence interval [CI], 1.03–1.05; P < 0.001). Each 1% increase in reticular opacity (HR, 1.09; 95% CI, 1.06–1.12; P < 0.001), ground-glass opacity (HR, 1.02; 95% CI, 1.01–1.04; P < 0.001), and honeycombing (HR, 1.12; 95% CI, 1.03–1.23; P = 0.010) was also associated with an increased hazard of death or lung transplant (Table 2). No outcome association was observed for change in vascular volume. Similar findings were observed in the validation cohort, but with less precise estimates because of the smaller sample size. When modeling the association between 2-year TFS and categorical change in qCT measures using optimal thresholds (Table E2), similar findings were observed (Table 2). Similar findings were also observed when modeling RMST (Table E3), with decreasing lung volume and increasing percentage reticular opacity and ground-glass opacity associated with significantly lower RMST across cohorts (Table E3).

Table 2.

Change in Quantitative Computed Tomography Measures and Association between Continuous and Categorical Change in Quantitative Computed Tomography Measures and 2-Year Transplant-Free Survival in University of Michigan Test and University of Chicago Validation Cohorts

	University of Michigan				University of Chicago
Continuous qCT measure (%)	Mean ± SD	HR*	95% CI	P Value	Mean ± SD	HR*	95% CI	P Value
Lung volume	−0.2 ± 18.0	1.04	1.03–1.05	<0.001	−1.8 ± 19.7	1.03	1.01–1.05	0.003
Vascular volume	4.4 ± 23.4	1.01	1.00–1.02	0.134	16.2 ± 42.4	1.02	1.01–1.03	<0.001
Reticular opacity	0.4 ± 5.8	1.09	1.06–1.12	<0.001	0.0 ± 5.9	1.07†	0.98–1.16	0.133
Ground-glass opacity	−0.2 ± 15.6	1.02	1.01–1.04	<0.001	3.9 ± 19.9	1.02	1.00–1.04	0.022
Honeycombing	0.2 ± 1.5	1.12	1.03–1.23	0.010	−0.1 ± 0.7	1.01	0.43–2.35	0.989

Categorical qCT measure (%)	“High” Strata, n (%)	HR*	95% CI	P Value	“High” Strata, n (%)	HR*	95% CI	P Value
Lung volume	105 (30)	3.20	2.07–4.94	<0.001	35 (38)	2.75	1.41–5.36	0.003
Vascular volume	286 (82)	2.25	1.15–4.38	0.018	82 (89)	0.88	0.31–2.47	0.802
Reticular opacity	45 (13)	2.87	1.76–4.68	<0.001	14 (15)	1.40	0.58–3.36	0.457
Ground-glass opacity	40 (11)	2.18	1.25–3.79	0.006	21 (23)	3.40	1.55–7.48	0.002
Honeycombing	94 (27)	1.90	1.24–2.92	0.003	12 (13)	1.28	0.45–3.62	0.646

Definition of abbreviations: CI = confidence interval; HR = hazard ratio; qCT = quantitative computed tomography.

All models adjusted for presence of connective tissue disease, sex, race, smoking history, baseline body mass index, age, percent predicted FVC, percent predicted Dl_CO, and CALIPER-derived measures total lung volume, pulmonary vascular-related structure volume, percent ground-glass opacity, reticular opacity, and honeycombing. Lung volume and vascular volume (pulmonary vascular-related structure volume) are reported as relative change, whereas reticular opacity, ground-glass opacity, and honeycombing are reported as absolute change.

^*Per unit change in qCT measures when modeled continuously and for “high” versus “low” strata after dichotomization at the optimal threshold.

^† Proportional hazards assumption violated.

Quantitative CT Measure of PPF

Model building results are shown in Table E4. A composite model composed of continuous change in lung volume, ground-glass opacity, and honeycombing was found to be the most discriminatory model, with a C-statistic of 0.66. Using this model, a qCT PPF risk score was determined using the following equation, which normalized the optimal threshold at a value of zero:

$$\begin{array}{l}\mathbf{\text{qctPPF}} \mathbf{\text{score}}=\left(.0088\ast \% \mathrm{change} \mathrm{in} \mathrm{ground} \mathrm{glass}\right)+\left(.1167\ast \% \mathrm{change} \mathrm{in} \mathrm{honeycombing}\right)-\left(.0285\ast \% \mathrm{change} \mathrm{in} \mathrm{relative} \mathrm{total} \mathrm{lung} \mathrm{volume}\right)-0.108\end{array}$$

After categorizing qctPPF scores above zero as qctPPF(+) and scores below zero as qctPPF(−), 45% (n = 256) and 52% (n = 48) of individuals were classified as qctPPF(+) in the test and validation cohorts, respectively. Figure E1 shows representative baseline and follow-up CT and CALIPER images of two qctPPF(+) cases. Those classified as qctPPF(+) had higher mean decline in lung volume and mean increase in ground-glass opacity, reticular opacity, and honeycombing when compared with those classified as qctPPF(−) (Figure E2). Among those classified as qctPPF(+), 96% experienced declining lung volume, 78% experienced increasing ground-glass and reticular opacities, and 53% experienced increasing honeycombing (Table E5). Correlation between changes in honeycombing, ground-glass opacity, and reticular opacity and lung volume at follow-up CT was weak (Figure E3).

Those classified as qctPPF(+) had significantly shorter 24-month TFS across cohorts (Figure 2). A qctPPF(+) classification was associated with a 4.4-fold increased hazard of death or lung transplant (HR, 4.41; 95% CI, 2.77–7.03; P < 0.001) in the test cohort and a 3.5-fold increased hazard of death or lung transplant (HR, 3.54; 95% CI, 1.62–7.71; P = 0.001) in the validation cohort. Similar results were again observed when modeling RMST, with qctPPF(+) classification associated with a 17.73% (95% CI, −23.56 to −11.89; P < 0.001) and 25.92% (95% CI, −41.96 to −9.88; P = 0.002) absolute decrease in RMST in the test and validation cohorts, respectively.

Figure 2.

Kaplan-Meier plots of 2-year transplant-free survival according to qctPPF classification in the (A) University of Michigan and (B) University of Chicago cohorts. qctPPF = quantitative computed tomography measurement of progressive pulmonary fibrosis.

In pooled cohort analysis, qctPPF(+) classification was associated with a 6.4-fold increased hazard of 24-month death or transplant among those with non-IPF ILD (n = 251) (HR, 6.40; 95% CI, 3.50–11.73; P < 0.001) and with a 3.1-fold increased risk in those with IPF (n = 191) (HR, 3.09; 95% CI, 1.78–5.36; P < 0.001) (Table 3). Formal interaction testing showed no evidence of effect modification by IPF diagnosis on the qctPPF TFS association (P_interaction = 0.16). When we substratified non-IPF ILD subtypes, those with CTD-ILD (n = 136) had a 5.6-fold increased risk of death or lung transplant (HR, 5.58; 95% CI, 2.22–13.99; P = 0.001), whereas those with other fibrotic ILD (IIP and fHP; n = 115) had a 8.5-fold increased risk of death or lung transplant (HR, 8.54; 95% CI, 3.38–21.61; P < 0.001). Formal interaction testing again showed no evidence of effect modification by ILD subtype (P_interaction = 0.34).

Table 3.

Adjusted Association between Quantitative Computed Tomography Measure of Progressive Pulmonary Fibrosis Status and 2-Year Transplant-Free Survival in Pooled Cohort, Stratified by Interstitial Lung Disease Subtype

Cohort (n)	qctPPF(+) (n/%)	HR	95% CI	P Value
Pooled (442)	204 (46.2)	4.28	2.90–6.31	<0.001
Non-IPF ILD (251)	110 (43.8)	6.40	3.50–11.73	<0.001
IPF (191)	94 (49.2)	3.09	1.78–5.36	<0.001
CTD-ILD (136)	61 (44.9)	5.58	2.22–13.99	0.001
Other fibrotic ILD (115)	49 (42.6)	8.54	3.38–21.61	<0.001

Definition of abbreviations: CTD = connective tissue disease; ILD = interstitial lung disease; IPF = idiopathic pulmonary fibrosis; qctPPF = quantitative computed tomography measurement of progressive pulmonary fibrosis.

All models adjusted for sex, race, smoking history, baseline body mass index, age, percent predicted FVC, percent predicted Dl_CO, and CALIPER-derived measures of total lung volume, pulmonary vascular-related structure volume, and percent ground-glass opacity, reticular opacity, and honeycombing.

Comparison with Radiology Report–determined PPF

Radiology report–determined PPF and qctPPF were concordant in 219 individuals and discordant in 131 (κ = 0.20), including 111 individuals who were qctPPF(+)/radPPF(−) and 20 who were qctPPF(−)/radPPF(+). Figure 3 shows TFS for individuals with PPF according to radPFF and qctPPF classification. qctPPF(−) individuals displayed the best overall TFS regardless of radPPF, whereas concordant positive individuals displayed the worst overall TFS. When comparing TFS between discordant individuals, the qctPPF(+)/radPPF(−) group had significantly worse survival than qctPPF(−)/radPPF(+) individuals (P = 0.03).

Figure 3.

Kaplan-Meier plots of 2-year transplant-free survival according to qctPPF and radPPF classification. qctPPF = quantitative computed tomography measurement of progressive pulmonary fibrosis; radPPF = radiologist-determined PPF.

Progression Analysis

In pooled cohort analysis of individuals with available longitudinal lung function measures after baseline CT (n = 359 for FVC; n = 348 for Dl_CO), mean 1-year change in percent predicted FVC after baseline CT was −6.07% (±9.54%) in the qctPPF(+) group (n = 157) and 0.76% (±8.34%) in the qctPPF(−) group (n = 202) (difference, 6.83; 95% CI, 4.97–8.69; P < 0.001). Mean 1-year change in percent predicted Dl_CO after baseline CT was −7.32% (±10.76%) in the qctPPF(+) group (n = 151) and −0.78% (±10.85%) in the qctPPF(−) group (n = 197) (difference, 6.54; 95% CI, 4.24–8.84; P < 0.001). Physiological PPF by either FVC or Dl_CO criteria was observed in 65.6% (n = 103) of individuals classified as qctPPF(+) and 37.6% (n = 76) of those classified as qctPPF(−) (P < 0.001). Those with both qctPPF and physiological PPF (pftPPF) displayed the worst 24-month TFS, whereas those with neither displayed the best TFS (Figure 4). Survival was similar for groups that displayed either qctPPF or pftPPF, suggesting equivalence when either was present.

Figure 4.

Kaplan-Meier plots of 2-year transplant-free survival according to qctPPF and pftPPF classification. qctPPF = quantitative computed tomography measurement of progressive pulmonary fibrosis; pftPPF = physiological PPF.

Among individuals with available longitudinal lung function measures after qctPPF classification (n = 300 for FVC; n = 288 for Dl_CO), mean 1-year change in percent predicted FVC was −2.76% (±7.70%) in the qctPPF(+) group (n = 134) and −1.82% (±7.99%) in the qctPPF(−) group (n = 166) (difference, 0.94%; 95% CI, −0.85%, 2.74%; P = 0.30) (Table E6). Mean 1-year change in Dl_CO percent predicted after qctPPF classification was −2.24% (±8.86%) in the qctPPF(+) group (n = 129) and −2.02% (±10.06%) in the qctPPF(−) group (n = 159) (difference, 0.22%; 95% CI, −2.00%, 2.45%; P = 0.84) (Table E6). In sensitivity analysis, higher penalties for missing data because of death or lung transplant resulted in larger differences between qctPPF strata (Table E6).

Discussion

In this study, we identified and validated several longitudinal qCT measures associated with 24-month TFS, with declining total lung volume and increasing ground-glass opacity demonstrating the strongest association with subsequent death or lung transplant across two independent cohorts. A novel qctPPF measure effectively discriminated TFS across independent cohorts and showed consistent measures of association across a diverse group of fibrotic ILDs, suggesting generalizability. This study adds to a growing body of work demonstrating that change in qCT measures could be used to classify PPF.

This work builds on recent studies evaluating the prognostic implications of longitudinal change in qCT in patients with ILD. Initial studies focused on IPF, showing that increasing qCT extent of fibrosis correlated with lung function decline over time (11, 14, 20, 27–30) and predicted subsequent survival (13, 19, 21). More recent studies have begun to focus on non-IPF ILD (18, 31, 32). Koh and colleagues recently applied texture analysis software (AVIEW Lung Texture) to 468 individuals in a non-IPF ILD cohort, showing that increasing qCT extent of fibrosis was associated with increased all-cause mortality (18). In a study by Ahn and colleagues of 97 individuals with fibrotic CTD-ILD and median follow-up of 30 months, PPF based on both visual assessment and qCT was an independent risk factor for decreased survival (32). Our findings reinforce this work and similar to a few of the mentioned studies (13, 21, 32), we employed a short 1-year time frame when deriving measures of qCT change, which approximates current recommendations for ILD monitoring based on limited studies in individuals with CTD-ILD (1, 33).

Our study also builds on prior studies aimed at establishing optimal quantitative thresholds when defining ILD progression. Humphries and colleagues found that a change in qCT fibrosis extent of 3% was the minimum clinically important difference in individuals with IPF when anchored to change in FVC (14). Ahn and colleagues found that a similar threshold was associated with CTD-ILD survival (32). To establish our qctPPF measure, we pursued a data-driven approach, selecting thresholds based on optimal outcome discrimination. Unsurprisingly, a composite measure of these variables better discriminated TFS than standalone measures, which was true across independent cohorts. Although our qctPPF measure effectively discriminates outcomes, the high baseline ground-glass opacity scores suggest that this label and other labels may not perfectly align to intended features. Although originally trained on radiologist-determined ground-glass opacity, our findings suggest that this algorithm may be detecting fibrotic features through the ground-glass opacity measure.

Individuals with qctPPF and physiological measures of PPF (FVC and Dl_CO decline) displayed the worst overall TFS, suggesting that a combination of qCT and physiological measures identifies a group at highest risk of poor clinical outcomes. Interestingly, those with either qctPPF or pftPPF also displayed reduced TFS, suggesting that satisfying either feature also has prognostic implications. We also showed that qctPPF performed similarly across diverse ILD subtypes. Compared with qctPPF(−) individuals, qctPPF(+) cases with IPF and non-IPF ILD had more than double the risk of death or transplant. Taken together, these findings suggest that our novel measure of qctPPF provides important prognostic information for diverse fibrotic ILD subtypes.

We also found that qctPPF classification did not predict subsequent lung function change. These findings corroborate existing literature for IPF and non-IPF ILD suggesting that baseline and longitudinal change in FVC do not reliably predict subsequent change in FVC (34, 35). Whether this reflects a treatment effect remains unclear, because treatment data were not available in these cohorts. Although antifibrotic therapy has been shown to slow FVC decline in patients with IPF (36), both antifibrotic and immunosuppressive agents are of potential benefit in those with non-IPF ILD (37).

Change in qCT-derived pulmonary vascular-related structure volume did not consistently predict TFS in this study, which stands in contrast to prior studies showing the prognostic significance of baseline measures of vascular volume (38). Alterations in the pulmonary vasculature can be complex in pulmonary fibrosis. Prior studies suggest that pulmonary vascular volume is increased in patients with both IPF (39) and other fibrotic ILD (40); however, other studies have shown that vascular pruning can lead to decreased lower total vascular volume and greater odds of interstitial lung abnormality progression in community-dwelling adults (41). These somewhat conflicting concepts require further attention.

Finally, comparisons between radiologist-determined PPF and qctPPF status revealed a lack of concordance. When qctPPF is compared with radiology reports of PPF, the kappa statistic (0.2) suggests that quantitative algorithms may be more sensitive to subtle changes in fibrosis than visual interpretation by a radiologist, because there were five times more cases of qctPPF(+)/radPPF(−) cases than qctPPF(−)/radPPF(+) cases. Those with the worst survival may have had more overt clinical progression that was recognized in both the radiology and the qctPPF measures. In the discordant radPPF and qctPPF pairings, Kaplan-Meier plots comparing survival (Figure 3) demonstrated worse outcomes in the qctPPF(+) individuals in which the radiology report did not describe PPF compared with the qctPPF(−) individuals with radiology reports indicating PPF. These findings support the potential use of qCT to detect earlier clinically meaningful evidence of PPF. Doing so could reduce the time to diagnosis, thereby speeding treatment initiation for this irreversible process.

Limitations

We acknowledge several limitations in this study. Given the retrospective nature of this study, chest CT scans were obtained in a nonstandardized way, which resulted in different protocols, slice thickness, and kernels being used for this study. Although all CT scans included satisfied the minimal quality requirement for CALIPER analysis, these differences in CT parameters likely introduced some degree of measurement error. However, this measurement error was unlikely to be different by outcome, reducing the likelihood of information bias. To some extent, poor inspiratory effort during the CT scan itself has the potential to impact changes in our qCT measures. Because our qCT measures required all individuals to have a baseline and at least one subsequent follow-up chest CT scan, there is also likely inherent selection bias in the patients undergoing repeat imaging, because these scans may have been prompted by a change in clinical status or decline in PFT results. A portion of patients were excluded from analysis in both the test and validation cohorts because of missing clinical data, which may reduce the representativeness of the studied cohorts. Next, the clinical radiology report was used to ascertain radiologist-reported progression, which could have introduced measurement error. Finally, patient-reported outcomes and treatment data were not available in this dataset to assess how much worsening respiratory symptoms, augmented immunosuppression, or antifibrotic treatment initiation could affect these models.

Conclusions

We found that a qCT-based measure of PPF predicted reduced survival across two independent ILD cohorts made up of diverse subtypes of fibrotic ILD. These findings provide further evidence that qCT measures may identify PPF earlier and more consistently than qualitative visual image interpretation, improving PPF classification and informing clinical decision making. Prospective validation of these findings with standardized high-resolution CT acquisition will be important as extensions of this work and should guide future iterations of PPF CT imaging criteria.