Multidimensional Clinical Deterioration in Interstitial Lung Disease: Longitudinal Concordance of Frailty, Patient-Reported Outcomes, and Lung Function in a Multinational European Registry Cohort Study

Abstract

Introduction

In interstitial lung disease (ILD), progression is tracked along lung physiology, patient-reported outcome measures (PROMs), and frailty, and prior registries report that these dimensions correlate at the group level. Whether deterioration on one axis identifies the same individual patients as another has not been examined.

Methods

In 371 adults from the multinational European ILD Registry (eurILDreg), spanning the full ILD spectrum, baseline-to-last-visit change was classified as decline at the minimal clinically important difference across eight instruments (CFS, MRC dyspnoea, K-BILD, LCQ, EQ-5D-5L, cough VAS, FVC %pred, DLCO %pred). Patient-level agreement was quantified by Cohen’s κ, interpreted direction-sensitively; a pre-specified window-matched analysis was primary, to account for differinginstrument follow-up windows differed (analysis-specific n 105–178). Outcomes (death or lung transplantation) were ascertained in 312 patients (24 events).

Results

At the individual-patient level, frailty, PROMs, and lung function deteriorated in largely different patients: No pair of the eight instruments exceeded κ = 0.33, and the discordance was not explained by measurement timing, diagnostic heterogeneity (persisting within IPF, κ 0.07–0.26), or a single domain – the two ILD PROMs themselves disagreed (K-BILD × LCQ κ = −0.19). Physiological (FVC-defined) progression was independently associated with K-BILD decline (OR 5.03). A spirometry-invisible subgroup declining on K-BILD while FVC remained stable (22% of 134) carried the highest event rate (13.3%).

Conclusion

In ILD, deterioration in frailty, patient-reported outcomes, and lung physiology identified largely non-overlapping populations at the level of the individual patient, with low agreement even within the patient-reported domain. Because the subgroup with declining patient-reported outcomes but preserved spirometry showed the highest event rate, assessment based on any single measure, and on FVC in particular, may not capture all deteriorating patients. These findings support the integration of validated patient-reported outcomes and frailty measures into multidimensional assessment and indicate that prospective validation of multi-axis decline against clinical outcomes should inform future ILD research and the design of progression criteria.

Introduction

Interstitial lung disease (ILD) encompasses a heterogeneous group of fibrosing and non-fibrotic conditions in which progression is monitored along multiple, partly independent clinical axes. International guidelines anchor disease behaviour in lung physiology, forced vital capacity (FVC), and diffusing capacity for carbon monoxide (DLCO), and the 2022 definition of progressive pulmonary fibrosis (PPF) requires at least two of three criteria (worsening symptoms, physiological progression, radiological progression) within the preceding year in a patient with an ILD other than idiopathic pulmonary fibrosis (IPF) [1–3]. In parallel, patient-reported outcome measures (PROMs) including the King’s Brief Interstitial Lung Disease (K-BILD) questionnaire, the Leicester Cough Questionnaire (LCQ), and the EQ-5D-5L are increasingly recommended for routine ILD assessment [4, 5], and the Clinical Frailty Scale (CFS) has emerged as an independent prognostic marker in ILD that captures global vulnerability not reducible to lung function [6, 7].

Pivotal randomised trials, reviews and registry data on antifibrotic therapy in IPF and progressive fibrosing ILD have reported parallel group-level changes in physiology and PROMs and concluded that the two move together [8–11]. The recent IPF-PRO Registry analysis of IPF patients [12] reported only weak correlations between continuous changes in lung function and changes in patient-reported outcomes; the INSIGHTS-IPF longitudinal data [11] showed concordant mean-level trajectories of K-BILD and FVC; the PROFILE cohort [13] demonstrated that cough burden tracks IPF progression only weakly. Frailty, assessed by the CFS and by a deficit-accumulation frailty index, has been associated with mortality in fibrotic ILD independent of FVC [7, 14], and the K-BILD has been validated as responsive in progressive fibrosing ILD [15]. Each of these studies has examined a single axis or a single instrument pair; none has quantified, at the patient level and at clinically interpretable thresholds, whether decline on frailty, on PROMs, and on physiology identifies the same or different patients over time.

This distinction matters operationally. If decline on each axis identifies the same patients, multidimensional assessment is redundant and single-axis monitoring is adequate; if it identifies distinct subsets, single-axis monitoring underestimates the disease burden it cannot see. We analysed the eurILDreg multinational ILD cohort to address this directly. The primary endpoint was the pairwise patient-level concordance, measured at minimal clinically important difference (MCID) thresholds, and expressed by Cohen’s κ, between frailty (CFS), the dominant ILD PROM (K-BILD), and the standard physiological instrument (FVC % pred). Secondary endpoints were the proportion of patients deteriorating on ≥2 of these three axes across underlying-diagnosis groups and across physiological (FVC-defined) progression, and the baseline and disease-behaviour predictors of decline on each axis. As exploratory analyses, we examined whether decline patterns differed by treatment exposure and how they related to outcomes (death and lung transplantation) ascertained from registry follow-up.

Methods

Study Design and Cohort

We analysed prospective longitudinal data from 371 consecutive ILD patients recruited between January 2021 and March 2026 in eurILDreg, a multinational prospective ILD registry coordinated by the Universities of Giessen and Marburg Lung Center (UGMLC) [16]. Inclusion required adult age (≥18 years), a multidisciplinary ILD diagnosis, and ≥2 visits with the same instrument separated by ≥6 months. Contributing centres were Giessen, Catania, Munich LMU, and Barcelona. The cohort size was set by the available registry cohort data. The registry enrolled both incident and prevalent patients, so baseline lung-function measurements for prevalent patients may predate registry enrolment; systematic patient-reported-outcome capture within the registry commenced in 2022. The study is reported in accordance with the STROBE guideline for observational studies, and the completed STROBE checklist is provided as online supplementary material (for all online suppl. material, see https://doi.org/10.1159/res/aepag007).

Diagnostic Classification

We classified patients by underlying ILD aetiology (IPF; systemic autoimmune rheumatic disease-associated ILD [SARD-ILD] including interstitial pneumonia with autoimmune features [IPAF]; hypersensitivity pneumonitis [HP], fibrotic and non-fibrotic; sarcoidosis; non-IPF idiopathic interstitial pneumonia [IIP], comprising nonspecific interstitial pneumonia, cryptogenic organising pneumonia, desquamative interstitial pneumonia, respiratory bronchiolitis-associated ILD, acute interstitial pneumonia, lymphocytic interstitial pneumonia, idiopathic pleuroparenchymal fibroelastosis, and unclassifiable IIP; rare ILD, comprising drug-induced, malignancy-associated, and eosinophilic ILD, storage diseases, familial pulmonary fibrosis, Langerhans cell histiocytosis, and other rare entities; and unclassifiable ILD) and modelled physiological progression as an orthogonal binary variable layered on the underlying non-IPF diagnosis. Physiological progression was operationalised as fulfilment of the FVC-based physiological criterion in a non-IPF patient; we use this term in preference to PPF because the 2022 ATS/ERS/JRS/ALAT definition of PPF is multidimensional and additionally requires symptomatic or radiological worsening that was not uniformly captured here [1].

Instruments and Scoring

K-BILD: 15 items scored 1–7, converted to 0–100, higher = better health status, MCID 5 [4]; LCQ: 19 items in three domains (physical, psychological, social), total = sum of three domain means (range 3–21, higher = better), MCID 1.3 [5]. EQ-5D-5L utility [17] was applied to the five-dimension levels, MCID 0.08 [18]. The Clinical Frailty Scale (CFS) [6] is a 9-point clinician-rated ordinal scale ranging from 1 (very fit) to 9 (terminally ill), MCID 1. The MRC dyspnoea scale [19] is a 5-point patient-reported ordinal scale of breathlessness, MCID 1. The Charlson Comorbidity Index (CCI) [20] is a weighted sum of comorbid conditions with an age adjustment. Patient-reported cough VAS was scored 0–100 with a 17-point cohort-internal change threshold. FVC %pred and DLCO %pred used the best manoeuvre per visit (GLI 2012 reference equations [21]) following ATS/ERS standards [22, 23]; MCID 5and 10 percentage points (absolute), respectively.

Treatment

ILD treatment was identified by ATC code from prospective registry capture and classified into three mutually exclusive categories for description. Antifibrotic-treated patients received nintedanib (L01EX09) or pirfenidone (L04AX05). Patients without antifibrotic treatment but with documented use of a systemic glucocorticoid (H02AB06/07: prednisolone) or a non-glucocorticoid immunosuppressant (mycophenolate mofetil L04AA06, azathioprine L04AX01, methotrexate L04AX03, rituximab L01FA01, cyclophosphamide L01AA01, hydroxychloroquine P01BA02, abatacept L04AA24, tocilizumab L04AC07, tacrolimus L04AD02) were classified as immunosuppressant treated. Remaining patients formed the reference group. Treatment exposure was defined as any documented use during the observation period (ever use), which does not distinguish transient courses, particularly of glucocorticoids, from chronic maintenance therapy. In the multivariable models, these exposures were instead entered as three overlapping binary covariates (antifibrotic, glucocorticoid, and non-glucocorticoid immunosuppressant), so that a patient receiving more than one contributed to each; treatment associations are observational and are not interpreted causally.

Statistical Analysis

Pairwise patient-level concordance was quantified by Cohen’s κ [24] with 2,000-iteration percentile-bootstrap 95% CIs. κ was interpreted direction-sensitively: values near zero indicate agreement no better than chance and negative values agreement below chance, and no absolute-value threshold was applied. Because per-instrument follow-up windows differed, the window-matched (common-window) analysis was pre-specified as the primary concordance comparison. Within-pair asymmetry of decline was tested by McNemar’s exact test in the patients with both measures; the 28-cell κ matrix was Benjamini-Hochberg FDR adjusted. A pre-specified sensitivity analysis repeated the primary pairs within IPF alone. Diagnostic-group and outcome comparisons used Fisher’s exact test, quadrant comparisons the χ² test, and the trend in event rates across the number of declining axes the Cochran-Armitage test; treatment-stratified changes were compared by Kruskal-Wallis test. Multi-axis decline (≥2 of CFS, K-BILD, FVC) was compared across underlying diagnosis and across progression category (physiological progression vs. IPF vs. non-progressive non-IPF) by Kruskal-Wallis test. Multivariable logistic regression was fitted for each decline outcome with covariates: underlying diagnosis, physiological progression, age, sex, baseline CFS, baseline CCI, baseline FVC %pred, and ILD treatment; a pre-specified sensitivity analysis additionally included the instrument’s inter-assessment interval. The CFS-decline model was reported descriptively. Outcome and treatment-stratified analyses were exploratory. Analyses used Python (statsmodels, scipy). Missing data were handled by complete-case analysis, each analysis using all patients with complete data for its variables without imputation; per-analysis denominators are reported throughout and in a participant-flow diagram (online suppl. Fig. 1). To assess whether missingness was informative, patients with versus without a longitudinal trajectory for each axis, and with versus without ascertained outcome status, were compared on baseline severity and mortality (online suppl. Table 4).

Ethics

The eurILDreg and eurIPFreg consortia hold ethics approvals from the Institutional Review Boards of all participating centres, including Giessen (AZ 111/08). All procedures were conducted in accordance with the Declaration of Helsinki; all patients provided written informed consent; the registries are registered at ClinicalTrials.gov (NCT02951416) and DRKS (DRKS00028968).

Results

Cohort and Baseline Characteristics

Of 371 patients, the underlying-diagnosis distribution was IPF 114 (31%), SARD-ILD (including IPAF) 73 (20%), HP 47 (13%), sarcoidosis 34 (9%), non-IPF IIP 26 (7%), rare ILD 38 (10%), and unclassifiable ILD 39 (11%) (Table 1). Forty patients (11%) met the physiological-progression criterion in addition to a non-IPF underlying diagnosis. Antifibrotic prescribing followed the regulatory indication: 80% of patients with physiological progression (32/40) and 68% of IPF patients (78/114) were antifibrotic treated. Among non-IPF patients, physiological progression by underlying diagnosis was HP 16, SARD-ILD 12, rare ILD 6, non-IPF IIP 4, unclassifiable 2, sarcoidosis 0. Concomitant immunosuppressant and antifibrotic use is detailed in online supplementary Table 1. The eligible cohort and the denominator for each analysis are shown in the participant-flow diagram (online suppl. Fig. 1); the number contributing to each analysis varied with the availability of paired measurements and of ≥2 visits ≥6 months apart, and exclusion reflected a single available visit or non-administration of that instrument. Missingness on the frailty and patient-reported axes was not associated with baseline severity or mortality, and outcome status was missing for 59 patients who differed only marginally from the remainder (older, higher CCI; all p ≥ 0.07); details are in online supplementary Table 4.

Table 1.

Baseline characteristics and per-instrument decline rates by underlying diagnostic group

Characteristic	IPF	SARD-ILD	HP	Sarcoid	non-IPF IIP	Rare ILD	uILD
N	114	73	47	34	26	38	39
Female, n (%)	24 (21)	34 (47)	19 (40)	15 (44)	4 (15)	14 (37)	21 (54)
Age, median (IQR), years	71 (64–76)	68 (59–76)	70 (63–75)	62 (45–69)	72 (64–79)	70 (63–77)	68 (57–74)
Baseline FVC %pred, median (IQR)	77 (62–87)	70 (60–82)	58 (47–73)	78 (69–91)	82 (75–93)	73 (55–84)	74 (57–77)
Baseline DLCO %pred, median (IQR)	45 (37–56)	44 (35–55)	39 (29–49)	64 (54–76)	47 (27–59)	59 (36–72)	58 (30–62)
CCI, median (IQR)	4 (3–6)	5 (3–6)	4 (3–5)	3 (2–5)	4 (4–6)	4 (2–6)	3 (2–5)
Baseline K-BILD, median (IQR)	70 (58–81)	74 (58–82)	68 (52–78)	71 (64–88)	70 (63–92)	69 (58–81)	61 (52–75)
Baseline CFS, median (IQR)	3 (3–5)	3 (3–4)	3 (3–5)	3 (2–4)	3 (3–3)	3 (2–4)	3 (3–3)
Physiological progression, n (%)	0 (0)	12 (16)	16 (34)	0 (0)	4 (15)	6 (16)	2 (5)
Antifibrotic ever, n (%)	78 (68)	27 (37)	21 (45)	1 (3)	6 (23)	20 (53)	9 (23)
Glucocorticoid ever, n (%)	15 (13)	39 (53)	28 (60)	28 (82)	9 (35)	4 (11)	4 (10)
Non-GC immunosuppr. ever, n (%)	4 (4)	44 (60)	10 (21)	7 (21)	4 (15)	0 (0)	1 (3)
Any immunosuppr. ever, n (%)	16 (14)	51 (70)	29 (62)	28 (82)	9 (35)	4 (11)	5 (13)
Per-instrument decline rates (k/n)
CFS	3/58	5/43	4/33	2/22	4/14	2/21	1/11
MRC	9/62	5/44	3/35	1/21	2/14	4/26	3/12
K-BILD	29/55	17/43	15/35	3/19	5/15	12/27	7/12
LCQ	5/54	9/43	7/35	5/19	3/15	4/28	2/12
EQ-5D-5L	26/56	20/44	9/35	7/19	6/15	10/28	4/12
Cough VAS	15/69	10/49	6/39	2/23	3/19	3/31	1/12
FVC %pred	31/60	15/40	9/26	11/26	3/6	4/18	6/9
DLCO %pred	22/61	15/40	7/26	8/26	1/6	7/18	5/9

Physiological progression is shown as an orthogonal row across diagnoses.

Per-Instrument Decline Rates

Decline rates by underlying diagnostic group are shown in Figure 1 and Table 1. Across the cohort, decline rates ranged from 10% (CFS) to 43% (K-BILD and FVC). Per-instrument follow-up windows differed substantially: the frailty and patient-reported axes were assessed over a shorter, more recent window (median 23.5–28.2 months) than lung function (FVC 52.2 months, DLCO 52.1 months) because systematic patient-reported-outcome capture began later in the registry. Concordance is therefore reported primarily from the window-matched analysis below.

Fig. 1.

Decline rates at the MCID by instrument and underlying diagnostic group. Cells show the percentage of patients declining at the instrument-specific MCID within each diagnostic group, shaded by decline rate. Per-instrument follow-up windows differ (frailty and patient-reported instruments: median 22–29 months; FVC and DLCO: median 51–55 months) and are not window-matched in this display. CFS, Clinical Frailty Scale; SARD-ILD, systemic autoimmune rheumatic disease-associated ILD (includes IPAF); DLCO %pred, diffusing capacity for carbon monoxide, percent predicted; EQ-5D-5L, EuroQol 5-Dimension 5-Level utility score; FVC %pred, forced vital capacity, percent predicted; HP, hypersensitivity pneumonitis; IPF, idiopathic pulmonary fibrosis; K-BILD, King’s Brief Interstitial Lung Disease questionnaire; LCQ, Leicester Cough Questionnaire; MCID, minimal clinically important difference; MRC, Medical Research Council dyspnoea scale; non-IPF IIP, non-IPF idiopathic interstitial pneumonia; VAS, visual analogue scale.

Primary Endpoint: Pairwise Concordance

Individual patients frequently deteriorated on one clinical axis without deteriorating on another. In the pre-specified window-matched primary analysis (each patient restricted to a common window across the three axes; median 21.6 months, n = 105–118), agreement was no better than chance: CFS × K-BILD κ = 0.10 (95% CI: −0.02 to 0.24), CFS × FVC κ = −0.07 (−0.19 to 0.04), and K-BILD × FVC κ = 0.28 (0.09–0.45). The all-available analysis gave concordant estimates (CFS × K-BILD κ = 0.07 [−0.04 to 0.18], n = 178; CFS × FVC κ = −0.04 [−0.15 to 0.09], n = 146; K-BILD × FVC κ = 0.18 [0.00 to 0.34], n = 134; Fig. 2). Interpreted direction-sensitively, all three primary values indicate agreement no better than chance, the negative CFS × FVC value indicating agreement below chance. Across the full eight-instrument matrix, no pair reached moderate agreement; the highest values, all κ ≈ 0.33, occurred between instruments sharing a measurement channel, namely, the two spirometry-derived measures (FVC × DLCO), two health-status questionnaires (K-BILD × EQ-5D-5L), and the two clinician-rated functional scales (CFS × MRC). The two ILD patient-reported outcomes did not agree with one another (K-BILD × LCQ κ = −0.19), and cough-related quality of life (LCQ) showed no positive agreement with any other axis (all κ ≤ 0.01, several below chance), so the discordance extends within the patient-reported domain rather than reflecting only a physiology-versus-PROM divide. Within-pair McNemar tests (FDR-adjusted) showed asymmetry between CFS and K-BILD or FVC (q < 0.0001) but not between K-BILD and FVC (q = 0.60), indicating that the low PROM-physiology concordance reflects different patients deteriorating on different axes rather than different decline rates. Restricting the primary pairs to IPF alone did not abolish the discordance (CFS × K-BILD κ = 0.11 [0.00 to 0.25]; CFS × FVC κ = 0.07 [−0.00 to 0.21]; K-BILD × FVC κ = 0.26 [−0.03 to 0.54]), all estimates remaining low; the discordance is therefore not an artefact of pooling heterogeneous diagnoses, though the smaller IPF subgroup widens the confidence intervals.

Fig. 2.

Pairwise patient-level Cohen’s κ for decline concordance across eight ILD instruments. Each cell shows the κ value with the patient-overlap n. The diverging colour scale is centred at κ = 0; κ is interpreted direction-sensitively, with values near zero indicating agreement no better than chance. Observed (uncorrected) concordance proportions per pair are provided in online supplementary Figure 2; these are high across pairs because decline is uncommon, which is why the chance-corrected κ shown here is the primary measure.

Sensitivity Analyses

Across the window-matched, all-available, and IPF-restricted specifications, agreement between the frailty, patient-reported, and physiological axes remained low, confirming that the discordance is not driven by the unequal follow-up windows or by diagnostic heterogeneity.

Multi-Axis Decline by Diagnosis and by Progression Category

Among 127 patients with complete follow-up on all three primary axes (CFS, K-BILD, FVC), the proportion declining on ≥2 axes did not differ significantly across the seven underlying diagnostic groups (Kruskal-Wallis p = 0.469; Fig. 3a). When classified by progression category, with physiological progression treated as orthogonal to underlying aetiology, multi-axis decline separated: physiological progression (non-IPF) 47.8% (11/23), IPF 38.5% (15/39), non-progressive non-IPF 16.9% (11/65); Kruskal-Wallis p = 0.002 (Fig. 3b). Only 4 patients (3.1%) declined on all three axes simultaneously, indicating that multi-axis decline typically presents as a partial rather than a global pattern.

Fig. 3.

Number of axes (CFS, K-BILD, FVC) on which patients met the MCID criterion for decline (n = 127 with complete three-axis follow-up). a By underlying diagnostic group, Kruskal-Wallis p = 0.469. b By progression category (physiological progression, IPF, non-progressive non-IPF), Kruskal-Wallis p = 0.002. The annotated value above each bar is the percentage declining on ≥2 axes.

Predictors of Decline Differ across Axes

Multivariable logistic regression (online suppl. Table 2; Fig. 4) identified largely non-overlapping predictors of decline on each axis; additionally adjusting for the inter-assessment interval left these associations materially unchanged, and the interval predicted decline on no axis (all p ≥ 0.25). Physiological progression independently correlated with K-BILD decline (OR 5.03 vs. no progression, 95% CI: 1.14–22.20, p = 0.033), with the IPF and SARD-ILD groups showing elevated but nonsignificant point estimates (OR 6.03 and 4.73). EQ-5D-5L decline was independently associated with age (OR 1.07 per year, p = 0.017); CCI did not retain significance after adjustment. FVC decline was associated with higher baseline FVC (OR 1.04, p = 0.001) and with antifibrotic treatment (OR 4.37, p = 0.009), the latter consistent with confounding by indication rather than a treatment effect. CFS decline was bounded by baseline CFS (OR 0.27 per point, p = 0.001) and lower baseline FVC (OR 0.95, p = 0.030); with only 17 events this model is descriptive. Glucocorticoid and non-glucocorticoid immunosuppressant exposures did not independently predict decline on any axis after adjustment.

Fig. 4.

Multivariable logistic regression forest plot. Adjusted ORs (log scale) with 95% CIs for each covariate as a predictor of decline on five binary outcomes (K-BILD, LCQ, EQ-5D-5L, FVC %pred, CFS). Coloured saturated points and thicker lines indicate p < 0.05; faded points are nonsignificant.

K-BILD × FVC quadrants

Among 134 patients with complete K-BILD and FVC trajectories (Fig. 5a), 22% (n = 30) declined on both, 22% (n = 30) on K-BILD only, 18% (n = 24) on FVC only, and 37% (n = 50) on neither. The both-decline quadrant was older (mean 70.8 years), male predominant (70%), heavily IPF (15/30), with the highest antifibrotic treatment (73%; Fig. 5d). The K-BILD-only quadrant was younger (mean 66 years), with the lowest baseline FVC (mean 67.5%pred; Fig. 5c) and the most diverse diagnostic composition with HP (n = 8), SARD-ILD (n = 7), and IPF (n = 6) the largest contributors (Fig. 5e). Antifibrotic treatment differed significantly across quadrants (χ²p = 0.007).

Fig. 5.

K-BILD × FVC trajectory quadrants (n = 134). a Scatter of change in K-BILD vs. change in FVC %predicted, coloured by quadrant; dashed lines mark MCID decline thresholds. b, c Per-quadrant distributions of age and baseline FVC. d Antifibrotic exposure per quadrant with χ² test. e Diagnostic-group composition per quadrant.

Outcomes by Decline Pattern

In an exploratory analysis, survival status was available for 312 patients (24 ascertained events: 19 deaths and 5 lung transplants); 1 further patient had an unknown outcome and was censored. The event rate rose with the number of declining axes (0/43, 3/47 [6.4%], 3/33 [9.1%], 1/4 [25%]; Cochran-Armitage trend p = 0.020; Fig. 6b). The K-BILD-only quadrant had the highest event rate of any quadrant (4/30, 13.3%; Fig. 6a), above both-decline (10%) and FVC-only (8.3%), while the neither quadrant had no events. Given only 24 events, these subgroup differences are hypothesis generating, and the association of isolated K-BILD decline with outcomes (Fisher exact p = 0.001) requires validation. Per-diagnostic-group event rates followed the expected ranking (IPF 11.8%, non-IPF IIP 8.3%, rare ILD 11.1%, unclassifiable 8.3%, SARD-ILD 4.7%, HP 4.3%, sarcoidosis 2.9%; Fig. 6c).

Fig. 6.

Exploratory outcome analysis (24 death or lung transplantation events among 312 patients with ascertained status; 1 further patient with an unknown outcome was censored). a Event rate by K-BILD × FVC quadrant (n = 134). b Event rate by the number of declining axes with the Cochran-Armitage trend test (n = 127). c Event rate by underlying diagnosis (n = 312). Bars are annotated with event counts (events/total).

Treatment-Stratified Trajectories

In an exploratory, unadjusted comparison, the cohort divided into three mutually exclusive treatment groups: antifibrotic-treated (n = 162, 44%), immunosuppressant-treated (n = 94, 25%), and untreated (n = 115, 31%). These groups differed substantially at baseline (online suppl. Table 3): antifibrotic-treated patients were older, had the lowest baseline DLCO (42% vs. 53% and 60%, p = 0.004), the highest CCI and CFS, and concentrated physiological progression (32/162 vs. 5/94 and 3/115). Against this severity imbalance, only the physiological axes differed between groups: FVC change (antifibrotic Δ−6.2, immunosuppressant Δ + 0.92, untreated Δ + 1.32; p = 0.009) and DLCO change (Δ − 6.6, Δ − 0.6, Δ + 0.02; p = 0.033). Cough-related quality of life improved similarly in antifibrotic-treated and untreated patients (LCQ Δ + 1.31 and Δ + 1.36) and less in immunosuppressant-treated patients (Δ + 0.13; p = 0.022), so the LCQ difference is not antifibrotic specific. The frailty and remaining patient-reported axes (K-BILD, EQ-5D-5L, CFS, MRC, cough VAS) did not differ across groups. Because treatment allocation was confounded by indication, disease severity, and follow-up duration, these differences cannot be interpreted as treatment effects.

Discussion

In this multinational eurILDreg cohort, deterioration in frailty, patient-reported outcomes, and lung function was largely non-overlapping at the level of the individual patient. Agreement between the three primary axes did not exceed that expected by chance in the pre-specified window-matched analysis, and across the eight instruments examined no pair reached moderate agreement, the highest observed value being κ = 0.33. These observations refine, rather than contradict, prior registry evidence. Analyses from the IPF-PRO [12] and INSIGHTS-IPF [10, 11] registries have reported that mean trajectories of lung function and patient-reported outcomes move in parallel; the present data indicate that concordance at the level of group means need not imply concordance at the level of the individual patient, on whom monitoring decisions are based.

Three potential explanations for the observed discordance were examined, and each was found insufficient. Unequal follow-up between instruments is unlikely to account for the finding since the window-matched analysis, in which each patient was restricted to a common observation window, was pre-specified as primary and yielded concordant estimates. Diagnostic heterogeneity is also an incomplete explanation as agreement remained low when the analysis was restricted to idiopathic pulmonary fibrosis alone (κ 0.07–0.26). Finally, the discordance was not confined to the distinction between physiological and patient-reported measures: agreement was low within the patient-reported domain itself, where the two disease-specific instruments (K-BILD and LCQ) did not concur (κ = −0.19) and cough-related quality of life showed no positive agreement with the remaining axes. Frailty, dyspnoea, cough, and health status therefore appear to capture distinct components of deterioration rather than a single underlying process.

The predictors of decline were similarly non-overlapping across axes, and no single baseline profile identified patients who deteriorated on all measures, providing internal support for multidimensional assessment. One directional association merits comment: physiological (FVC-defined) progression was independently associated with subsequent K-BILD decline (OR 5.03) but not with FVC decline, which may indicate that the patient-reported axis registers deterioration in patients meeting a physiological progression criterion before further physiological loss becomes apparent. These findings are consistent with prior work. Baseline frailty has been associated with mortality in fibrotic ILD independently of lung function [7, 14]; the present analysis extends this observation to the longitudinal dimension by demonstrating that change in frailty is largely independent of change in physiology and in health status. The weak relationship between cough burden and disease progression is consistent with Saunders et al. [13], and the responsiveness of the K-BILD in progressive fibrosing ILD reported by Birring et al. [15] is corroborated at the individual-patient level. The observation that no single instrument captures all dimensions of ILD progression, noted by Wijsenbeek et al. [9], is here quantified, and the proposal by Behr et al. [25] to extend progression assessment beyond pulmonary function testing is supported.

These results have implications for disease monitoring. If deterioration is registered by different instruments in different patients, assessment based on a single axis will not capture patients deteriorating on the others. In an exploratory outcome analysis, patients whose K-BILD declined while FVC remained stable had the highest event rate in the cohort. Current definitions of ILD progression and the primary endpoints of antifibrotic trials are anchored predominantly in FVC; the present data suggest that, at the individual-patient level and across the ILD spectrum, such criteria may not identify a subset of deteriorating patients with a less favourable prognosis. Taken together, these observations provide a rationale for incorporating validated patient-reported outcomes and frailty measures into routine ILD assessment alongside lung function and identify a subgroup, namely, patients declining on patient-reported outcomes with preserved spirometry, that warrants prospective evaluation.

The treatment-stratified analysis was exploratory and subject to confounding by indication. Antifibrotic-treated patients, who were more physiologically advanced at baseline, declined on FVC and DLCO while their frailty and patient-reported trajectories did not change in parallel, a pattern that mirrors the axis discordance observed across the cohort (Fig. 7). The improvement in cough-related quality of life was not specific to antifibrotic therapy, being of comparable magnitude in untreated patients, and no between-group difference outside the physiological measures was evident given the substantial baseline imbalance. These comparisons cannot support causal inference and are not interpreted as treatment effects.

Fig. 7.

Exploratory treatment-stratified trajectories across the eight instruments, by mutually exclusive treatment group (antifibrotic n = 162, immunosuppressant n = 94, untreated n = 115). Each panel shows the mean baseline-to-last-visit change per group with the Kruskal-Wallis p value; significant between-group differences (marked with an asterisk) were confined to FVC, DLCO, and LCQ. Comparisons are unadjusted and confounded by indication. AF, antifibrotic; IS, immunosuppressant; Un, untreated.

The principal strength of this study is its design. Eight instruments spanning frailty, dyspnoea, disease-specific and generic health status, cough, and lung physiology were assessed prospectively in the same patients across the full ILD spectrum, permitting patient-level concordance to be examined across instrument classes in a manner not addressed by studies confined to one or two axes. The consistency of the discordance across the window-matched, all-available, and IPF-restricted analyses reduces the likelihood that it reflects measurement timing or case mix, and the K-BILD × FVC quadrant framework offers a clinically interpretable representation of decline patterns.

Several limitations should be considered. Analysis-specific denominators varied because each analysis required paired measurements of its constituent instruments; although the missingness analysis indicated that missingness on the patient-reported and frailty axes was not associated with baseline severity or mortality, informative missingness cannot be excluded, and outcome status was unavailable for 59 patients. Continuous change was dichotomised at instrument-specific thresholds, with attendant loss of information, and the choice of threshold influences the classification of decline. The outcome analysis was based on 24 events, and the prognostic observations, including those relating to the subgroup with isolated K-BILD decline, are therefore exploratory and hypothesis generating and require prospective validation. Radiological progression, a component of the multidimensional definition of PPF [1], was not captured. The treatment comparisons are observational and confounded by indication. The within-IPF sensitivity analysis was based on a small subsample with correspondingly wide confidence intervals, and the cohort was drawn from four European tertiary referral centres, which may limit generalisability to primary-care and non-European settings.

Conclusion

In ILD, deterioration in frailty, patient-reported outcomes, and lung physiology identified largely non-overlapping populations at the level of the individual patient, with low agreement even within the patient-reported domain. Because subgroup with declining patient-reported outcomes but preserved spirometry showed the highest event rate, assessment based on any single measure, and on FVC in particular, may not capture all deteriorating patients. These findings support the integration of validated patient-reported outcomes and frailty measures into multidimensional assessment and ILD healthcare.

Acknowledgments

We thank the European Joint Programme on Rare Diseases (RARE-ILD), the BMBF, and the German Center for Lung Research, all patients and physicians who contributed to the eurIPFreg/eurILDreg registry, and the staff of the Centre for Interstitial and Rare Lung Diseases at the Universities of Giessen and Marburg Lung Center for ongoing clinical and data-management support. The authors used DeepL (DeepL SE, www.deepl.com) as a translation aid for refining selected passages originally drafted in German. All translated text was reviewed, edited, and verified by the authors, who retain full responsibility for the content.

Statement of Ethics

This study was conducted in accordance with the principles of the Declaration of Helsinki. The eurIPFreg/eurILDreg study protocol was reviewed and approved by the Ethics Committee of the Justus-Liebig University Giessen, Approval No. 111/08. Written informed consent to participate in the study was obtained from all participants. eurIPFreg is registered at ClinicalTrials.gov (NCT02951416) and eurILDreg in the German Clinical Trials Register (DRKS00028968).

Conflict of Interest Statement

The authors have no conflicts of interest to declare.

Funding Sources

The eurILDreg is an independently governed, non-profit registry operating under the RARE-ILD consortium. It evolved from the eurIPFreg, originally established in 2008 within the European IPF Network under the European Commission’s Seventh Framework Programme. No specific external funding was received for the present analysis. The eurIPF-/eurILDreg consortium has received registry-maintenance support from Roche and Boehringer Ingelheim; neither company had any role in the design, analysis, interpretation, or drafting of the present manuscript.

Author Contributions

S.T. and E.K. conceived the study. A.C.W., H.M., and S.T. extracted the clinical and laboratory data. E.K. and A.C.W. performed and supervised the statistical analyses. S.T. and E.K. drafted the manuscript. P.M., J.B., B.C., C.V., M.M.M., and A.G. contributed to data interpretation and critically revised the manuscript for important intellectual content. All authors approved the final version of the manuscript.

Funding Statement

The eurILDreg is an independently governed, non-profit registry operating under the RARE-ILD consortium. It evolved from the eurIPFreg, originally established in 2008 within the European IPF Network under the European Commission’s Seventh Framework Programme. No specific external funding was received for the present analysis. The eurIPF-/eurILDreg consortium has received registry-maintenance support from Roche and Boehringer Ingelheim; neither company had any role in the design, analysis, interpretation, or drafting of the present manuscript.

Data Availability Statement

The data that support the findings of this study are not publicly available owing to privacy and registry-governance restrictions but are available from the corresponding author, and following approval by the eurILDreg Steering Committee, to qualified researchers upon reasonable request.