Relationship between high-resolution computed tomography quantitative imaging analysis and physiological and clinical features in antisynthetase syndrome-related interstitial lung disease

Abstract

Objectives

To explore the association between the extent of CT abnormalities by quantitative imaging analysis (QIA) and clinical/physiological disease parameters in patients with antisynthetase syndrome associated interstitial lung disease (ARS-ILD).

Methods

We analysed 20 patients with antisynthetase antibodies and active ILD enrolled in the Abatacept in Myositis-Associated Interstitial Lung Disease study. High-resolution chest CT was obtained at weeks 0, 24 and 48 and QIA scored the extent of ground glass (quantitative score for ground glass), fibrosis (quantitative score for lung fibrosis, QLF) and total ILD (quantitative ILD, QILD). Mixed-effects models estimated longitudinal QIA scores over time. Associations between QIA scores with clinical/physiological parameters were analysed longitudinally using repeated-measures mixed-effects models.

Results

Patients were median age 57 years, 55% males and 85% white. Higher (worse) baseline QIA scores correlated with lower baseline forced vital capacity (FVC) and diffusing capacity adjusted for haemoglobin (DLCO). Longitudinal QIA trajectories trended towards improving scores during the trial, and patients on O₂ at baseline had worsening QIA trajectories which were different from patients who were not on O₂. Longitudinal QIA scores demonstrated strong associations with both FVC and DLCO over time. Higher QILD scores over time were also associated with worse dyspnoea scores, pulmonary visual analogue scale, physician and patient global disease activity, health status in 6/8 domains of the Short Form-36 and higher oxygen requirements. Patients with significant radiographic improvement at 48 weeks had higher baseline QLF, QILD and worse DLCO.

Conclusions

Longitudinal QIA scores associate with lung physiology, patient perception of respiratory status, overall disease activity and quality of life over time in ARS-ILD. QIA may allow reproducible monitoring of disease progression and response to therapy over time.

Trial registration number

NCT03215927.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Interstitial lung disease (ILD) is a prominent feature of antisynthetase syndrome, and high-resolution chest CT plays an important role in the clinical evaluation of ILD. Computer-aided quantitative image analysis (QIA) provides reproducible quantification of ILD-related parenchymal abnormalities and has been used as a surrogate outcome measure in clinical trials of scleroderma-associated ILD and idiopathic pulmonary fibrosis.

WHAT THIS STUDY ADDS

• Our results show that QIA scores over time have moderate to strong correlations with lung physiology as well as clinical outcome measures of respiratory status, quality of life and physician and patient perception of the disease in anti-aminoacyl-tRNA synthetase-related ILD (ARS-ILD).

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• QIA has the potential to serve as an imaging biomarker that empowers objective and reliable monitoring of disease progression and response to therapy in ARS-ILD.

Introduction

Idiopathic inflammatory myopathies (IIMs, collectively known as myositis) are a heterogenous group of disorders characterised by autoimmune-mediated inflammation primarily of the skeletal muscle along with a wide spectrum of systemic involvement. In patients with IIM with autoantibodies against aminoacyl-tRNA synthetases (ARS), interstitial lung disease (ILD) can be the first and most prominent feature¹ and is an important prognostic factor associated with poor survival.^{2 3} Earlier diagnosis and effective management of ILD are vital in improving the prognosis in ARS-ILD.¹ Despite increased awareness of the significance of ARS-ILD, only a few clinical trials have been performed in this population, probably due, in part, to the paucity of reproducible objective respiratory measures of which the application has been studied in this patient group.

High-resolution chest CT (HRCT) plays an important role in the evaluation of ILD and is a good surrogate marker of important histological findings with prognostic implications in scleroderma-related ILD (SSc-ILD), rheumatoid arthritis-related ILD (RA-ILD) and idiopathic pulmonary fibrosis (IPF).⁴,⁹ There has been increasing interest in using quantitative imaging analysis (QIA) for computer-aided quantification of the parenchymal abnormalities of ILD as a surrogate outcome measure in clinical trials of SSc-ILD and IPF.¹⁰,¹⁴ The complexity and nature of parenchymal abnormalities seen in myositis-ILD and ARS-ILD are different from those seen in SSc-ILD and other connective tissue disease-related ILD and the applicability in this patient group is largely unknown.

In a recent retrospective observational study from a South Korean myositis-ILD cohort, cross-sectional QIA scores correlated well with lung physiological measures, and higher baseline QIA scores were predictive of higher risk of lung transplantation or death.¹⁵ However, longitudinal associations between QIA scores and physiological measures over time were not reported in this study. Also, the association between extent of parenchymal abnormalities on HRCT and patient-reported outcomes of dyspnoea, health-related quality of life (QOL) and overall disease activity have not been assessed in myositis-ILD or ARS-ILD.

In the current study, we explored the association between longitudinal quantitative CT scores using QIA and comprehensive physiological and patient-centred indices of disease over time in patients with active ARS-ILD from a 48-week multicentre randomised interventional study.

Methods

Patient selection

We analysed 20 patients enrolled in the Abatacept in Myositis-Associated Interstitial Lung Disease (Attack My-ILD) study (ClinicalTrials.gov identifier: NCT03215927). All subjects gave written informed consent for the study. Patients with antisynthetase antibodies and active ILD were enrolled across four centres in the USA. Antisynthetase antibodies were tested by commercial assays available at the patient’s local laboratory which typically used ELISA. Enrolled patients were randomised to a double-blind placebo period for 24 weeks assigned to either abatacept (n=9) or placebo (n=11), followed by an open label period in which all patients were on abatacept for 24 weeks. All patients were on standard of care immunosuppressive agents for ILD (glucocorticoid or steroid-sparing agent, either azathioprine or mycophenolate, or combination of steroid and one steroid-sparing agent) prior to the trial and were maintained on a stable standard of care regimen and a taper of glucocorticoids throughout the trial.

Active ILD was defined as new-onset ILD (within 3 months of ILD diagnosis) or recent worsening or flare-up of existing ILD requiring treatment, as defined by one or more of the following: (a) worsening of ground glass opacities, reticulation, honeycombing or fibrosis on chest HRCT as per local thoracic radiologist or pulmonologist within the previous 6 months, (b) relative decline in forced vital capacity (FVC) ≥10% within 3 months and/or (c) new oxygen requirement within 6 months. Criteria for active ILD were reviewed by an adjudication committee prior to enrolment. Patients with FVC≤30% predicted, on supplemental O₂≥6 L/min for ≥1 month, listed for lung transplantation or with chronic stable ILD or end-stage fibrosis in which the investigator determined there was low potential for improvement were excluded from the trial.

HRCT image assessment by QIA and visual analysis

Chest HRCT scans obtained with standard non-contrast thin section protocols at each institution were analysed. HRCTs were performed at three timepoints (weeks 0, 24 and 48) during the trial and pre-baseline scans were also included when available. Image quality was assessed, and scans obtained with contrast protocol, >3 mm slice thickness, poor resolution or artefacts were excluded. Bilateral lungs (right, left) were each segmented into three zones (upper—apices to aortic arch, middle—aortic arch to inferior pulmonary veins, lower—below inferior pulmonary veins). We applied computer-aided QIA using previously published methods.¹⁶,¹⁸ QIA scores were assigned as a percentage of the extent of ground glass (quantitative score for ground glass, QGG), fibrotic patterns (quantitative score for lung fibrosis, QLF), honeycombing (quantitative score for honeycombing, QHC) and consolidations (quantitative score for consolidation, QCON). The quantitative ILD (QILD) score represents the sum of QGG+QLF+QHC+QCON. QIA scores were also reported as the absolute volume of the portion of the lung with abnormal parenchyma adjusted for the total lung volume (QIA in mL), in order to account for different levels of inspiration that may vary even within the same patient. QIA scores were calculated for whole lung and also for each zone. The zone of maximum involvement (ZM) was defined as the zone with the highest QILD score at baseline. Radiographic improvement in ILD was defined as QILD decline (improvement) of ≥2% of the whole lung, which was used as the anchor for minimal clinically important difference in SSc-ILD.^{19 20}

Two thoracic radiologists (CC, FA) who were blinded to the patient’s clinical data visually assessed the CT images, which included overlays of the QILD as quantified by QIA. They confirmed that the computer-aided quantification matched the visually assessed extent of ILD with over 80% agreement. Additionally, the radiologists conducted a consensus review to identify the predominant CT features and visual ILD patterns using Fleischner Society glossary of terms.^{21 22} The readers determined fibrosis to be present if there was evidence for reticulation with architectural distortion with or without traction bronchiectasis and regional volume loss.

Pulmonary function test and clinical outcome measures

Pulmonary function tests (PFTs) and the clinical outcome measures were obtained at five timepoints (weeks 0, 12, 24, 36 and 48). Clinical respiratory outcome measures included dyspnoea score from the University of California San Diego Shortness of Breath Questionnaire (UCSD SOBQ dyspnoea),²³ pulmonary visual analogue scale (VAS) from the Myositis Disease Activity Assessment Tool,²⁴ supplemental oxygen (O₂) use and a 6 min walk test (6MWT) that measures 6 min walk distance (6MWD)²⁵ as well as pretest and post-test Borg scale for dyspnoea and fatigue.²⁶

Myositis outcomes were also obtained at these visits including a physician global myositis disease activity VAS (MD global VAS 0–10), extramuscular VAS, manual muscle testing in eight muscle groups (MMT, 0–150) and patient-reported global myositis activity VAS (patient global VAS).²⁴ QOL measures included Short Form-36 (SF-36)^{27 28} and Health Assessment Questionnaire-Disability Index (HAQ-DI).²⁴

PFTs were obtained at each institution using standardised protocols to obtain FVC and diffusing capacity adjusted for haemoglobin (DLCO) in accordance with ATS/ERS recommended standards.²⁹,³² Both measures were reported as % predicted values as well as by absolute volume (mL for FVC, mL/min/Hg for DLCO). Per cent predicted values were calculated centrally using the age, height and haemoglobin collected from the trial using the global lung function initiative equations.^{33 34}

Statistical analysis

Baseline associations between quantitative CT scores with various physiological and clinical parameters were analysed cross-sectionally using Spearman’s correlations.

Mixed-effects models were used to estimate longitudinal trajectories of whole lung QIA scores (in % and in mL) over time. We also tested interactions with treatment group, age, sex, baseline supplemental O₂ use, baseline QIA scores, dyspnoea score, FVC and DLCO to determine whether CT score trajectories were different in patient subgroups. In patients who had pre-baseline scans, we compared mixed-effects model slopes of pre-baseline QIA score trajectories to trajectories during the trial (post-baseline) by treatment group using piecewise linear splines.

Repeated-measures analysis using mixed-effects linear models was performed to analyse the longitudinal associations between CT scores and physiological/clinical measures across multiple timepoints. The results, presented as standardised regression coefficients (β), indicate the expected change in the dependent variable in terms of SDs per unit change in the predictor, facilitating a comparison of the relative importance of the predictors.³⁵ Fixed-effects models estimated the associations over time within individuals.³⁶ Association strengths were classified as weak (<0.2), moderate (0.2–0.59) or strong (≥0.6). Interactions with age, baseline O₂ use and muscle strength (MMT<150 vs MMT=150) were also tested. Changes in QIA from baseline were explored in relation to clinical parameter changes using unstandardised regression coefficients to estimate effect sizes. Responders to therapy were determined at 48 weeks based on radiographic ILD improvement from baseline. Baseline characteristics of responders were compared with non-responders using Student’s t-test or Wilcoxon rank-sum test for continuous variables and χ² test for categorical variables.

All statistical testing was two sided with a significance threshold of 0.05 except for exploratory analysis testing associations between changes from baseline in which p value threshold was 0.10 in view of the smaller number of follow-up scans. Statistical analysis was performed using JMP Pro V.16.0 (SAS Institute, Cary, North Carolina, USA) and SAS V.9.4 (SAS Institute).

Patient and public involvement

There was no patient or public involvement in the design or conduct of the study.

Results

Cross-sectional analysis at baseline visit

Patients were middle aged (median age 57 years), 55% male and 85% white with median ILD duration of 2 years (table 1). Anti-Jo1 antibody was the most common antisynthetase antibody (55%) and remaining subjects had two anti-PL-7 and one of each anti-PL-12, OJ and EJ antibodies. Nine (45%) patients had muscle involvement and MD global disease activity was moderate (VAS 4.0 (1.6–5.5), median (IQR)).

Table 1. Baseline characteristics.

Clinical variable	Total (n=20)	Placebo (n=11)	Abatacept (n=9)
Age, years	56.7 (47.2–60.5)	57.7 (47.8–64.5)	49.7 (46.6–59.3)
Female sex	9 (45)	4 (36)	5 (56)
Non-Hispanic ethnicity	20 (100)	11 (100)	9 (100)
Race
White	17 (85)	10 (91)	7 (78)
Black	2 (10)	1 (9)	1 (11)
Asian	1 (5)	0	1 (11)
Anti-Jo1 ab (vs non-Jo1 ab)	11 (55)	7 (64)	4 (44)
ILD duration, months	25 (5–51)	17 (3–52)	39 (6–67)
Height, cm	173 (160–180)	178 (160–183)	165 (158–177)
Weight, kg	89 (75–102)	90 (81–103)	83 (65–101)
BMI, kg/m2	28.4 (25.5–35.7)	31.7 (25.5–35.7)	27.8 (23.8–37.0)
Supplemental O2 use, yes	6 (30)	3 (27)	3 (33)
PFT
FVC, % predicted	64 (52–73)	61 (53–78)	66 (49–69)
DLCO Hb, % predicted	48 (40–59)	49 (41–64)	47 (38–52)
HRCT analysis, n*	17†	10†	7
Visual ILD patterns, n (%)
Mixed NSIP	4 (24)	4	0
Fibrotic NSIP	3 (18)	2	1
Cellular NSIP	1 (6)	0	1
Fibrotic OP	3 (18)	1	2
OP	1 (6)	1	0
Indeterminate for UIP	4 (24)	2	2
Other	1 (6)	0	1
Predominant CT feature
Ground glass	5 (29)	3 (30)	2 (29)
Fibrosis	8 (47)	5 (50)	3 (43)
Consolidation	3 (18)	2 (20)	1 (14)
Other	1 (6)	0	1 (14)
HRCT scores with QIA
Whole lung QIA, %
QILD-WL	38.6 (27.9–50.7)	36.3 (28.3–45)	43.7 (26.3–61.8)
QGG-WL	19.1 (14.3–27.3)	18.9 (14.9–22.5)	19.6 (13.2–29.3)
QLF-WL	16.2 (12.1–23.9)	17.0 (13.4–22.4)	14.3 (10.3–38.4)
QHC-WL	0 (0–0.6)	0.2 (0–0.9)	0 (0–0.4)
QCON-WL	0.2 (0.1–0.5)	0.2 (0.1–0.4)	0.2 (0.1–0.8)
Zone of maximum involvement, n (%)
RUZ	2 (12)	1 (10)	1 (14)
LUZ	4 (24)	2 (20)	2 (29)
LMZ	1 (6)	0	1 (14)
RLZ	3 (18)	2 (20)	1 (14)
LLZ	7 (41)	5 (50)	2 (29)
Zone of maximum involvement QIA, %
QILD-ZM	78.2 (69.3–86.1)	83.3 (75.2–87.2)	76.2 (62.7–85.9)
QGG-ZM	17.3 (13.6–27.1)	16.2 (14.1–21.1)	25.7 (8.2–29.8)
QLF-ZM	52.9 (44.6–72.9)	63.5 (52.5–71.9)	47.0 (41.2–74.9)
QHC-ZM	0 (0–0.4)	0.05 (0–0.9)	0 (0–0)
QCON-ZM	0.3 (0.1–1.9)	0.3 (0.1–1.4)	0.5 (0.2–2.8)

Values are reported as median [(interquartile rangeIQR]) or n (%).

^*Visual reads performed only on scans that met quality control standards for QIA (n=17).

^†One patient did not have baseline scans during the trial but had CT scans from 6 months before trial entry. FVC and DLCO at pre-baseline scan and baseline visit were similar.

abantibodiesBMIbody mass indexDLCOdiffusing capacity adjusted for haemoglobinFVCforced vital capacityHRCThigh-resolution chest CTILDinterstitial lung diseaseLLZ, left lower zoneLMZleft middle zoneLUZ, left upper zone; NSIPnon-specific interstitial pneumoniaOPorganising pneumoniaPFTpulmonary function testQCONquantitative score for consolidationQGGquantitative score for ground glassQHCquantitative score for honeycombingQIAquantitative image analysisQILDquantitative ILDQLFquantitative score for lung fibrosisRLZ, right lower zone; RUZ, right upper zoneUIPUsual interstitial pneumoniaWLwhole lungZMzone of maximum involvement

All patients had active ILD at baseline with moderate to severe physiological impairment based on a median FVC of 64% predicted and DLCO 48% predicted, and six (30%) patients were on supplemental O₂. Baseline CT characteristics by visual analysis and QIA are presented in table 1. The most common ILD pattern was non-specific interstitial pneumonia (NSIP)-type pattern (47%), and fibrosis was the predominant CT feature in eight (47%) patients. On QIA, median QILD at baseline was 39% of the whole lung, which was mostly ground glass (QGG 19%) and fibrotic patterns (QLF 16%), while honeycombing (QHC 0%) and consolidation (QCON 0.2%) were absent or minimal. Maximum involvement (ZM) was most frequently seen in lower lung zones.

Spearman’s correlation of baseline characteristics with baseline CT scores showed that higher (worse) QILD scores had moderate to strong correlations with worse baseline FVC and DLCO (r=−0.51 to −0.64, p<0.05). Higher QGG correlated with worse FVC, and higher QLF correlated with worse DLCO. Higher QLF also correlated with shorter 6MWD at baseline (r=−0.54, p=0.03). Baseline QGG and QILD also correlated with patient global VAS (r=0.56 and 0.59, p=0.02 for both), but not with MD global VAS or extramuscular VAS. Worse health status on SF-36 domains for physical functioning, emotional role functioning, bodily pain and general health perceptions also correlated with higher baseline QIA scores cross-sectionally (online supplemental table 1 for details).

Longitudinal quantitative CT scores during the trial period

All baseline and follow-up CT scans that were adequate for QIA were analysed and the slope of the QIA trajectories during the trial period was estimated using mixed-effects linear models (figure 1 for QIA %, online supplemental figure 1 for QIA in mL adjusted for total lung volume). Estimated trajectories for QIA scores over time in the entire cohort trended towards overall improvement.

Figure 1. Quantitative CT scores (QIA-WL % scores) over time in all patients with HRCT scans (n=17). Bold lines are estimated trajectories of QIA (QGG, QLF, QILD) % whole lung scores over time (months) using mixed-effects models in 17 patients who had one or more HRCT scans adequate for QIA. Mean change±SE for QGG was −0.12±0.09 per month, p=0.21; for QLF −0.19±0.18 per month, p=0.30; for QILD −0.33±0.24 per month, p=0.19. HRCT, high-resolution chest CT; QGG, quantitative score for ground glass; QIA, quantitative image analysis; QILD, quantitative interstitial lung disease; QLF, quantitative score for lung fibrosis; WL, whole lung.

Longitudinal QIA scores over time were also estimated in patient subgroups to test whether QIA trajectories differ by certain baseline characteristics. Baseline O₂ use was a significant effect modifier in which patients on O₂ at baseline trended towards worsening QLF scores over time while patients not on O₂ trended towards improved QLF scores over time (interaction p=0.04 in figure 2a for QIA %, p=0.03 in online supplemental figure 2a for QIA in mL). Patients with baseline organising pneumonia (OP)-type pattern trended towards greater improvement in QLF and QILD scores compared with non-OP-type patterns although the interaction was not statistically significant (figure 2b). Longitudinal QIA scores over time also did not have interactions with the following variables measured at baseline: age, sex, ILD disease duration, dyspnoea score, FVC or DLCO (interaction p value=NS for all).

Figure 2. Longitudinal quantitative CT scores in different patient subgroups. P value of interaction term. Estimated trajectories of QIA (QGG, QLF, QILD) % whole lung scores over time (months) in patient subgroups by (a) baseline supplemental oxygen use and (b) baseline visual ILD pattern. ILD, interstitial lung disease; NSIP, non-specific interstitial pneumonia; OP, organising pneumonia; QGG, quantitative score for ground glass; QIA, quantitative image analysis; QILD, quantitative ILD; QLF, quantitative score for lung fibrosis.

Longitudinal quantitative CT scores correlate with PFTs

Mixed-effects linear models were used to estimate the association between QIA scores and PFTs over all timepoints (table 2). Whole lung QGG, QLF and QILD in % all demonstrated moderate to strong associations with both FVC and DLCO (β=−0.35 to −0.64, p<0.05 for all). QIA in mL had similar associations with FVC and DLCO % predicted (online supplemental table 2), while QIA-ZM scores did not (online supplemental table 3). Fixed-effects models for within-person correlations showed DLCO had strong associations with QLF and QILD, while FVC associations with QLF and QILD were no longer statistically significant (online supplemental table 4).

Table 2. Associations between quantitative CT scores and clinical/physiological parameters over time using mixed-effects models.

	QGG		QLF		QILD
	β	P value	β	P value	β	P value
FVC, % predicted	−0.64*	<0.01	−0.48*	0.01	−0.61*	<0.01
FVC, mL	−0.56*	<0.01	−0.41*	0.02	−0.54*	<0.01
DLCO Hg, % predicted	−0.35*	0.05	−0.51*	<0.01	−0.50*	<0.01
DLCO Hg, mL/min/mm Hg	−0.42*	0.02	−0.56*	<0.01	−0.57*	<0.01
6MWD, m	0.05	0.82	0.05	0.81	0.06	0.75
UCSD SOBQ dyspnoea	0.12	0.51	0.45*	0.02	0.37*	0.049
Pulmonary VAS	0.26	0.07	0.30	0.07	0.37*	0.02
Borg dyspnoea, pretest	0.13	0.53	−0.01	0.96	0.05	0.81
Borg fatigue, pretest	−0.13	0.51	−0.16	0.44	−0.16	0.44
Borg dyspnoea, post-test	0.14	0.46	0.23	0.22	0.24	0.22
Borg fatigue, post-test	0.01	0.96	0.23	0.26	0.19	0.37
Supplemental O2, L	0.41	0.06	0.59*	<0.01	0.61*	<0.01
MD global activity VAS, 0–10	0.29	0.08	0.47*	0.01	0.48	0.01
Extramuscular activity VAS, 0–10	0.28	0.08	0.38*	0.04	0.41	0.02
MMT8, 0–150	−0.03	0.87	0.10	0.56	0.06	0.73
Patient global VAS	0.29	0.06	0.40*	0.03	0.45*	0.01
HAQ-DI	0.08	0.62	0.10	0.63	0.09	0.65
SF-36
Emotional well-being	−0.15	0.32	−0.37*	0.04	−0.33	0.05
Emotional role functioning	−0.28*	0.046	−0.32	0.06	−0.37*	0.02
Vitality	−0.31*	0.049	−0.39*	0.03	−0.41*	0.02
General health perceptions	−0.56*	<0.01	−0.39	0.05	−0.53*	0.01
Bodily pain	−0.13	0.27	−0.24	0.11	−0.26	0.08
Physical functioning	−0.25	0.23	−0.30	0.16	−0.31	0.15
Physical health	−0.29	0.08	−0.36	0.06	−0.38*	0.04
Social role functioning	−0.34*	0.03	−0.45*	0.02	−0.46*	0.01

Estimated in 19 subjects who had one or more HRCT scans with QIA. $\mathit{\beta }$ are standardized regression coefficients using linear mixed-effects models.QIA scores in % of whole lung.

β is standardised regression coefficient using linear mixed-effects models.

*P<0.05. Bold values indicate results that are statistically significant (p<0.05).

QIA scores in % of whole lung.

DLCOdiffusing capacity adjusted for haemoglobinFVCforced vital capacityHAQ-DIHealth Assessment Questionnaire-Disability IndexHRCThigh-resolution chest CTMMT8manual muscle testing in eight muscle groups6MWD6 min walk distanceQGGquantitative score for ground glassQIAquantitative image analysisQILDquantitative ILDQLFquantitative score for lung fibrosisUCSD SOBQUniversity of California San Diego Shortness of Breath QuestionnaireVASvisual analogue scale

In order to estimate the relationship between change in QIA from baseline (visit QIA-baseline QIA) with change in PFT from baseline (visit PFT-baseline PFT), we performed mixed-effects models in patients who had a baseline and at least one follow-up QIA score. Improvement in QIA scores from baseline trended towards associations with improvement in FVC and DLCO from baseline (table 3). Each unit improvement in DLCO % predicted from baseline associated with 0.34 unit improvement from baseline in QLF (p=0.06), and 0.74 unit improvement from baseline in QILD (p=0.08).

Table 3. Association between changes in QIA whole lung scores from baseline and changes in clinical/physiological parameters from baseline.

Change from baseline	Change in QGG %		Change in QLF %		Change in QILD %
	Estimate	P value	Estimate	P value	Estimate	P value
FVC per % predicted change	−0.12	0.55	−0.69	0.17	−0.63	0.35
FVC per 100 mL change	−0.17	0.73	−1.66	0.18	−1.51	0.35
DLCO Hg per % predicted change	−0.14	0.33	−0.34 *	0.06	−0.74 *	0.08
DLCO Hg per mL/min/mm Hg change	−0.63	0.32	−1.49 *	0.06	−3.22 *	0.08
6MWD per 100 m change	0.03	0.99	−0.74	0.86	−0.85	0.88
UCSD SOBQ dyspnoea per unit change	0.01	0.91	0.22	0.26	0.27	0.33
Pulmonary VAS per unit change (0–10)	−0.09	0.87	1.75	0.17	1.52	0.38
Borg dyspnoea, pretest, per unit change (0–10)	2.04	0.48	0.33	0.94	2.26	0.75
Borg fatigue, pretest	0.30†	0.80	−1.34†	0.68	−1.67	0.68
Borg dyspnoea, post-test	0.26	0.78	0.92	0.63	1.08	0.69
Borg fatigue, post-test	0.39†	0.68	−0.04†	0.99	−0.36	0.91
HAQ-DI per unit change (0–3)	2.11	0.46	2.38	0.69	4.19	0.61
Patient global VAS per unit change (0–10)	0.07	0.90	2.18*	0.03	2.03	0.28
SF-36 domains per unit change (0–100)
Emotional well-being	−0.05	0.53	−0.43	0.10	−0.51	0.12
Emotional role functioning	−0.01	0.64	−0.04	0.38	−0.04	0.60
Vitality	−0.13	0.17	−0.30	0.17	−0.49	0.12
General health perceptions	−0.22	0.13	−0.40*	0.05	−0.85*	0.07
Bodily pain	−0.04	0.43	−0.08	0.43	−0.15	0.29
Physical functioning	−0.11	0.28	0.04	0.82	−0.11	0.70
Physical health	−0.04	0.37	−0.06	0.25	−0.14	0.25
Social role functioning	−0.07	0.36	−0.21*	0.05	−0.44*	0.08

Mixed -effects linear models in 16 patients with baseline and at least 1one follow-up HRCT that was adequate for QIA.

Estimate represents unstandardizsed regression coefficients from linear mixed -effects model estimating the relationship between absolute change from baseline in CT scores and absolute change from baseline in clinical/physiological parameters. Each unit change in predictor to each unit change in CT score.

^*Value reported after removing 1 that was an influential outlier.P<0.10.

^†Value reported after removing one patient that was an influential outlier.

DLCOdiffusing capacity adjusted for haemoglobinFVCforced vital capacityHAQ-DIHealth Assessment Questionnaire-Disability IndexHRCThigh-resolution chest CT6MWD6 min walk distanceQGGquantitative score for ground glassQIAquantitative image analysisQILDquantitative ILDQLFquantitative score for lung fibrosisSF-36Short Form-36UCSD SOBQUniversity of California San Diego Shortness of Breath QuestionnaireVASvisual analogue scale

In order to test whether associations between QIA scores and PFTs differed by certain baseline characteristics, we tested interactions with age, baseline O₂ use and muscle weakness (MMT) (online supplemental table 5). None of the interactions tested were statistically significant. However, the associations between QIA scores and PFTs were stronger in patients older than 50 (compared with those younger than 50) and in patients with normal muscle strength (compared with patients with MMT<150).

Longitudinal quantitative CT scores correlate with clinical respiratory outcomes and supplemental O₂ use

We also analysed the associations of QIA scores (by % and mL) with clinical respiratory outcomes over all timepoints (table 2 for QIA %, online supplemental table 2 for QIA in mL, online supplemental table 3 for QIA-ZM). Higher whole lung QLF and QILD scores had moderate associations with worse UCSD dyspnoea scores and higher O₂ use over time (by % and by mL). Higher QILD scores were also associated with worse pulmonary VAS over time. QIA scores were not associated with 6MWD or Borg scales of fatigue or dyspnoea over time.

To test whether clinical respiratory outcomes had stronger associations with QIA scores or with PFTs, we constructed mixed-effects models between PFT measures and the same respiratory outcomes (online supplemental table 6), and compared standardised coefficients to standardised coefficients in table 2. The UCSD dyspnoea score and pulmonary VAS had stronger coefficients with QIA scores than with PFT measures (for UCSD dyspnoea β=0.45 with QILD vs β=−0.23 to −0.31 with FVC and DLCO; for pulmonary VAS β=0.37 with QILD vs β=−0.18 to −0.27 with FVC and DLCO). O₂ use had strong positive correlations with QLF and QILD but was not associated with PFTs (β=−0.14 to −0.36, p=NS for all). On the other hand, 6MWD and Borg scores for dyspnoea and fatigue had many weak but significant associations with PFTs, but no significant associations with QIA scores. It is noteworthy that the trial design included more PFT observations than HRCT observations.

Longitudinal quantitative CT scores correlate with global disease activity and patient-reported outcomes of QOL

Higher longitudinal QILD and QLF scores of whole lung and ZM had moderate associations with higher MD global VAS, extramuscular VAS and patient global VAS over time, but not with MMT (table 2, online supplemental tables 2 and 3). Whole lung QIA scores also associated with worse health status by 6/8 domains of the SF-36 over time, while HAQ-DI did not (table 2).

The associations with whole lung QILD (table 2) were stronger than associations with PFT (online supplemental table 6) for MD global VAS, extramuscular VAS, patient global VAS and the aforementioned 6/8 SF-36 domains.

Responders with improvement in QILD at 12 months

12 patients had HRCT scans at baseline and 48 weeks. In these patients, the median (IQR) change in QILD from baseline to 48 weeks was −4% (−13, 13). Seven (58%) of these patients had improvement in QILD by >2% and were grouped as ‘responders’ (QILD change from baseline −11 (−30, –6), median (IQR)).

We compared baseline predictors between responders and non-responders (table 4). Treatment assignment was not significantly different between the two groups. Responders were more often females, had lower baseline DLCO and higher baseline QLF and QILD. All responders had full muscle strength at baseline which was significantly different from non-responders (median MMT 141). Among those with predominant CT features of fibrosis at baseline, 40% were responders compared with 71% of patients with predominantly non-fibrotic baseline CT features who became to be responders (p=0.27). Baseline QGG, FVC, visual ILD pattern type, age, disease duration, Jo1 versus non-Jo1, 6MWD, pulmonary VAS, Borg scores, UCSD dyspnoea, O₂ use, MD global activity VAS, HAQ-DI and SF-36 were not different between responders and non-responders.

Table 4. Baseline characteristics of patients who responded to treatment based on improvement in QILD by >2% at 48 weeks.

	Non-responder (n=5)	Responder (n=7)	P value
Treatment group, abatacept	2 (40%)	3 (43%)	0.92
QILD change from baseline	4 (3–12)	−11 (−30 to −6)	0.01
Age	51 (49–60)	56 (46–60)	0.68
Sex, female	2(40%)	6(86%)	0.048
ILD duration at baseline, months	39 (6–77)	17 (11–46)	0.81
Anti-Jo1 ab	4 (67%)	2 (33%)	0.56
Non-Jo1 ab	3 (50%)	3 (50%)
FVC, mL	2.6 (2.0–3.8)	2.2 (2.2–2.4)	0.46
FVC, % predicted	71 (60–78)	68 (57–73)	0.51
DLCO Hg, mL/min/mm Hg	13.7(12.4–17.8)	10.8(8.5–12.2)	0.04
DLCO Hg, % predicted	60 (52–66)	49 (40–55)	0.12
Baseline visual ILD pattern
NSIP type	3 (43%)	4 (57%)	0.92
OP type	1 (33%)	2 (67%)	0.73
Other	1 (50%)	1 (50%)	0.79
Baseline visually predominant CT feature
Fibrosis	3 (60%)	2 (40%)	0.27
Ground glass or consolidation	2 (29%)	5 (71%)
Baseline QGG	17 (11–24)	23 (18–29)	0.14
Baseline QLF	11(9–17)	25(21–39)	0.01
Baseline QILD	27(23–41)	53(41–63)	0.03
6MWD, m	386 (336–543)	324 (289–466)	0.16
Pulmonary VAS, 0–10	4.5 (2.8–6.4)	4 (2.5–5.6)	0.57
Borg dyspnoea, pretest	0.5 (0.25–3)	0 (0–0.5)	0.14
Borg dyspnoea, post-test	2 (2–3.5)	2.5 (0.88–3)	0.57
Borg fatigue, pretest	1 (0.5–2)	0 (0–0.5)	0.12
Borg fatigue, post-test	2 (0.5–3.5)	2 (0.5–2.25)	0.71
UCSD SOBQ dyspnoea	33 (19–52)	42 (29–58)	0.81
O2 use, L	0 (0–1.5)	0 (0–0)	0.90
MD global activity VAS, 0–10	5.5 (2.7–6.2)	5 (2.6–5.7)	0.37
Extramuscular activity VAS, 0–10	4 (3–5.8)	4 (1.8–6.5)	0.81
MMT8, 0–150	141(134–149)	150(150–150)	0.02
Patient global VAS, 0–10	3 (2.25–5.5)	5 (2.5–7)	0.46
HAQ, 0–3	0.25 (0.065–0.51)	0.63 (0.25–1)	0.08
SF-36 domains
Physical functioning	45 (33–78)	35 (20–40)	0.08
Physical health	25 (0–100)	0 (0–50)	0.54
Emotional role functioning	100 (17–100)	33 (0–100)	0.34
Vitality	50 (43–80)	50 (45–55)	0.74
Emotional well-being	80 (54–92)	64 (52–80)	0.41
Social role functioning	62 (50–88)	62 (50–75)	0.56
Bodily pain	67 (39–84)	67 (22–67)	0.67
General health perceptions	40 (25–100)	30 (25–35)	0.35

Values are n (%) or median [(IQR]). P value by sStudent’s t-test or Wilcoxon rank-sum test for continuous variables and chi-squareχ2 test for categorical variables.

Responders to therapy were determined at 48 weeks based on radiographic ILD improvement (QILD improvement ≥$\ge$2% of the whole lung) from baseline.

abantibodiesDLCOdiffusing capacity adjusted for haemoglobinFVCforced vital capacityHAQHealth Assessment QuestionnaireILDinterstitial lung diseaseMMT8manual muscle testing in eight muscle groups6MWD6 min walk distanceNSIPnon-specific interstitial pneumoniaOPorganising pneumoniaQGGquantitative score for ground glassQILDquantitative ILDQLFquantitative score for lung fibrosisSF-36Short Form-36UCSD SOBQUniversity of California San Diego Shortness of Breath QuestionnaireVASvisual analogue scale

Discussion

In the current work, we used a computer-aided QIA to measure the extent of ILD on CT images of patients with ARS-ILD to assess ILD severity over time, and to explore the relationship between QIA scores and clinically important functional and patient-reported outcome measures longitudinally. QIA scores demonstrated moderate to strong associations with both FVC and DLCO at baseline and over time. Higher QLF and QILD scores over time also associated with worse scores in multiple patient and physician-reported outcomes, including the UCSD dyspnoea score, pulmonary VAS, O₂ use, MD global activity VAS, extramuscular VAS, patient global VAS and 6/8 domains of the SF-36 longitudinally. Moreover, these outcome measures had more robust associations with quantitative CT scores than with FVC or DLCO.

Higher whole lung QILD scores had moderate to strong correlations with worse FVC and DLCO in ARS-ILD. Correlations between QILD with PFTs were seen in a cross-sectional analysis within an observational myositis-ILD cohort from Korea.¹⁵ Our study also evaluated longitudinal associations, demonstrating that changes in QIA scores associated with changes in physiological measures over time, reflecting responsiveness of QILD. Although QIA has mostly been validated in SSc-ILD,^{16 37} our findings support that this tool may also be useful in myositis-ILD including ARS-ILD as an outcome measure.

Among specific ILD features, QLF scores correlated strongly with DLCO, more so than with FVC or total lung capacity, a pattern also observed in SSc-ILD where DLCO is the best predictor of fibrosis extent.³⁷ In contrast, QGG scores showed stronger correlations with FVC. Ground glass opacities, while potentially indicating active inflammatory alveolitis, can also represent fine intralobular fibrosis or oedema in SSc-ILD.^{4 38} Longitudinal HRCT studies in SSc-ILD have shown that ground glass opacities may evolve into normal or fibrotic patterns, suggesting a decrease in QGG does not always indicate ILD improvement.¹⁰ In patients with NSIP, reductions in ground glass opacities correlated with improvements in FVC, more than in DLCO,³⁹ similar to our findings. Future studies should explore the mapping of specific CT features and their temporal changes to enhance our understanding of their clinical implications in ARS-ILD.

Quantitative CT scores allowed for group-level analysis and comparisons which would have been difficult with visual assessment. We found that patients on O₂ at baseline had worsening CT scores over time, unlike patients who were not on O₂, suggesting that patients on O₂ require closer monitoring and consideration for more aggressive interventions.

Physiological variables are indirect surrogates for the extent of structural disease and can be influenced by patient effort, gender and disease-related factors. In ARS-ILD, the presence of muscle weakness or pain, respiratory muscle weakness and diaphragmatic dysfunction may be additional factors that can impact PFT results. In our current work, PFTs had significant correlations with QIA scores in patients with full muscle strength (MMT=150), whereas correlations were weaker and mostly non-significant in patients with muscle weakness (MMT<150). Although FVC and DLCO are frequently used as a main outcome measure in ILD clinical trials, CT scores may offer a more direct and precise measure of the underlying structural pathological process.

In addition to its correlations with physiological measures, it is important to assess whether CT quantitation is providing a meaningful representation of physician and patient perception of the disease.²⁸ In our current work, higher QILD scores over time correlated with worse respiratory status longitudinally by UCSD dyspnoea scores, pulmonary VAS and higher O₂ requirement. The UCSD SOBQ measures the severity of dyspnoea during activities of daily living and has been validated in patients with IPF and fibrotic ILD.^{40 41} It is noteworthy that UCSD dyspnoea score and pulmonary VAS had more robust associations with QIA scores than with PFT measures. O₂ use was not associated with PFTs but strongly correlated with QLF and QILD scores. Our work suggests that QIA scores reflect patient and physician perception of dyspnoea and respiratory status, perhaps more closely than PFTs.

On the other hand, 6MWD and Borg scores at rest and after exertion with a 6MWT were not correlated with QIA scores. A possible explanation for the discrepant results between the 6MWT and other clinical respiratory outcome measures may be due to the fact that many factors other than parenchymal lung disease are known to impact the 6MWD, such as age, body habitus, sex and comorbidities that cause pain or limitations in mobility, as well as muscle weakness in myositis.⁴² While Borg scales and 6MWT may be valuable measures of patient function and perception of respiratory impairments, our sample size may have been insufficient to adjust for confounders and detect significant relationships between QIA and these measures.

Higher QILD and QLF scores were also associated with higher global disease activity and worse QOL by the SF-36 over time. Association of QIA scores with measures of the patient’s perception of overall disease and health-related QOL adds to its value in measuring disease that impacts daily living.

Patients who had significant improvement in radiographic ILD at 48 weeks had higher baseline QLF, QILD and worse DLCO. Female sex and absence of muscle weakness also appeared to be associated with a favourable outcome. The use of a reliable, reproducible quantitation tool can be helpful in identifying prognostic factors in ILD. For example, higher baseline QLF scores correlated with a favourable treatment response in SSc-ILD⁹ and high baseline QLF scores were associated with higher 5-year mortality in RA-ILD.⁴³ QIA can detect subtle progression of disease or a favourable response that may not be obvious with semiquantitative visual assessment, an advantage that can be particularly helpful for comparing cohorts that are already on background standard of care therapy.

Such advantages of QIA offer a mean for reliably following patients over time in clinic. Its high reproducibility and sensitivity to change can be helpful when assessing response to treatment and short-term changes can allow physicians to pinpoint patients with rapid/acute progression. QIA also has the theoretical advantage over PFTs or patient-reported outcome measures as it evaluates the extent of parenchymal lung disease with being minimally affected by patient effort, muscle weakness or musculoskeletal pain.

The current work has limitations. First, our sample size was small, mostly white and had a follow-up period limited to 48 weeks. IIM is a rare disease,⁴⁴ and the Attack My-ILD was a proof-of-concept study with strict inclusion criteria to only include patients with active ILD. The smaller sample size was collected from four centres in the USA and may not be representative of other populations. Also, the sample size and shorter follow-up time limited our ability to perform multivariate adjustments or robust evaluation of predictors of long-term outcomes and the significance level was not adjusted for multiple testing given the exploratory nature of the study. However, our comprehensive assessment of the repeated measures over time and their associations with CT scores over the same time period were made possible by using a well-phenotyped clinical trial cohort. While observational datasets and convenience cohorts often have the advantage of a larger sample size, they pose problems in relation to the inherent heterogeneity in patient characteristics, treatments during the follow-up period and variable timing of outcome assessments.

Another limitation of our study is the lack of standardisation in HRCT protocols such as use of non-volumetric versus volumetric scanners, slice thickness of images and ensuing breath-holds for maximum inspiration. We attempted to minimise variability in the data by performing within-person models and adjusting QIA scores for the total lung volume at each visit, but standardisation of HRCT protocols would be needed for optimal QIA scores.

Myositis-ILD and ARS-ILD have morphological patterns that may not be fully captured by the current QIA software, which was largely designed for SSc-ILD.^{16 17} For instance, whereas a previous study noted air-space consolidations in 48% of patients with ARS-ILD⁴⁵ and visual assessment in our cohort described consolidation as the predominant CT feature in 18% of scans, QIA identified a median QCON score of <1%, indicating potential limitations of the QIA in distinguishing consolidations from other patterns like fibrosis. QCON score was not included in our results and we postulate that changes in QLF values may in part represent improvement in consolidations. Nevertheless, visual assessments confirmed that the QILD score effectively differentiates and quantifies all ILD-related abnormalities in ARS-ILD. Also, the strong correlation of QILD with clinical measures underscores its utility. While QIA has been applied across various ILD conditions and in clinical trials,¹¹,^{1546 47} future work will aim to optimise the QIA algorithm for myositis-ILD and validate its use in larger prospective cohorts.

Summary and conclusions

In this initial exploratory study, we demonstrate that the extent of lung parenchymal abnormalities on chest HRCT using QIA associates with impairments in lung physiology as well as patient and physician-reported outcomes of respiratory status over time in patients with ARS-ILD. QIA has the potential to serve as an objective imaging biomarker in myositis-ILD and ARS-ILD that may allow objective and reliable monitoring of disease progression and response to therapy over time. Further work is needed to optimise and validate its use in myositis-ILD, determine whether specific CT features correspond to symptoms and/or treatment response and identify potential baseline predictors of long-term ILD outcomes.

Data availability statement

Data are available upon reasonable request.