Morphological computed tomography phenotypes in pulmonary sarcoidosis at diagnosis: splitting and lumping based on predictors of clinical progression Authors: Prerana Agarwal, Louis Jouanjan, Charlott Terschluse, Martin Feineis, Sebastian Wirtz, Daniel Soriano, Paula Kuss, Emil Wachenfeld, Fabian Bamberg, Christopher L. Schlett, Daiana Stolz, Björn C. Frye Published: 2026 ABSTRACT BackgroundSarcoidosis may represent a broad umbrella term for various diseases that induce non-necrotising granulomas in affected organs. A recent multinational consensus proposed distinct computed tomography (CT) phenotypes for pulmonary sarcoidosis; however, the clinical implications remain unclear. What is the prevalence of individual phenotypes in a tertiary referral cohort, and are these phenotypes associated with disease progression?Materials and methodsThis retrospective study comprised 232 biopsy-proven pulmonary sarcoidosis patients treated between 2010 and 2020. Baseline CT scans were grouped into 1) non-fibrotic/fibrotic/unclear for fibrosis, and 2) normal/nodular/non-nodular. Disease progression was assessed over 36 months using predefined clinical, functional and therapeutic criteria.ResultsMost patients had non-fibrotic phenotypes (91.0%, n=211), while fibrotic (5.6%, n=13) and unclear (3.4%, n=8) patterns were less frequent. Inter-reader agreement was moderate for individual phenotypes (κ=0.55) and good for fibrotic versus non-fibrotic classification (κ=0.79). At 36 months, 69% of patients had progressed; all fibrotic cases progressed, compared with 67% of non-fibrotic. Nodular phenotypes progressed in a similar way to normal CT (64% versus 68%) and at a lower rate than non-nodular disease (88%). Fibrotic disease was associated with an increased risk of disease progression after adjustment for age, sex, body mass index and baseline lung function (hazard ratio 2.31, 95% CI 1.21–4.38; p=0.038).ConclusionsCT-based phenotype assessment of sarcoidosis seems feasible but may require refinement. Fibrotic patterns predict progression, while nodular phenotypes carry a risk similar to normal CT. Further radiological splitting may not reflect distinct disease behaviour. Introduction Sarcoidosis is a multisystem, granulomatous disease of unclear aetiology, primarily affecting the lungs and intrathoracic lymph nodes [1–3]. Given its manifold clinical and radiological presentations, current discussions suggest that sarcoidosis may represent a broad umbrella term for various diseases that induce a non-caseating granulomatous response in the affected organs [4]. The clinical course of pulmonary sarcoidosis remains unpredictable, ranging from spontaneous resolution to a chronic disease course. Imaging plays an important role in the diagnosis and management of affected patients, and various efforts have been made to identify relevant imaging markers in sarcoidosis to predict disease behaviour [5–9]. While chest radiography has traditionally been the initial imaging modality used when pulmonary sarcoidosis is suspected, consensus is growing in favour of routine baseline evaluation with high-resolution computed tomography (HRCT). This shift is being driven by HRCT's superior inter-observer agreement, enhanced diagnostic and prognostic accuracy, and greater sensitivity in detecting clinically significant findings, such as fibrosis [10]. A recent multinational initiative has proposed a morphological CT-based classification of pulmonary phenotypes in sarcoidosis, developed using a modified Delphi process [4]. This effort marks a significant step towards facilitating future research aimed at unravelling the complexities of this disease as a basis for the possible definition of separate diseases. One of the primary objectives was to transition from a generalised “lumping” of sarcoidosis cases to a more refined “splitting” approach through CT-based classification. Consensus was reached on the classification of sarcoidosis into seven phenotypes that could be broadly categorised into non-fibrotic and possibly fibrotic patterns. The prevalence of the individual CT phenotypes in different populations has not yet been examined. Additionally, studies are needed that explore patterns that correlate with lung function, bronchoalveolar lavage findings, putative triggers and disease progression, which will contribute to a more precise characterisation of sarcoidosis. The aim of this study was to elucidate the prevalence of and characterise these morphological CT-phenotypes in our tertiary referral centre. Furthermore, we aimed to find groups of patterns that were more likely to be predictive of progression during a follow-up period of 36 months. Materials and methods Study design and study population The study population of this retrospective single-centre study comprised patients with pulmonary sarcoidosis who were treated at our tertiary care centre between 2010 and 2020. Inclusion criteria were biopsy-proven sarcoidosis in accordance with the joint statement of ATS/ERS/WASOG [11], with at least one CT scan at the time of baseline clinical and lung functional testing at our institute. Exclusion criteria were missing lung function data, poor quality of CT scans and patients whose baseline CT scan was more than 1 year apart from the lung function examination. The study was approved by the Institutional Research Ethics Board of the Medical Faculty of the Albert-Ludwig-University Freiburg (24-1284-S1-retro). The requirement for informed consent was waived by the Ethics Committee of the Medical Faculty of the Albert-Ludwig-University Freiburg owing to the study's retrospective nature. The requirements of the Helsinki Declaration on human research were met. CT examination As our centre is a sarcoidosis referral centre, many patients had baseline CT scans performed elsewhere. These scans were included if the image quality was satisfactory; examinations with a slice thickness of > 5 mm and poor image quality due to artefacts or insufficient coverage were excluded. CT scans were categorised based on predominant phenotypes as per the Delphi consensus statement. Briefly, seven different phenotypes were described in this statement: diffuse peribronchovascular micronodules, larger peribronchovascular nodules, scattered larger nodules unrelated to the bronchovascular structures, predominant consolidation, progressive massive fibrosis-like pattern, bronchocentric reticulation and upper lobe fibrocavitary disease. Further details are provided in Appendix A. Inter-reader variability A second reader (EW, radiology trainee, 3 years’ experience) performed additional readings of the entire sample (N=232) after being trained by the first reader. The inter-reader variability for the assessment of the phenotypes showed moderate agreement, with a Cohen's kappa of 0.55 (95% confidence interval (CI) 0.48–0.62; p <0.001), and good agreement for the assessment of fibrotic/non-fibrotic pattern, with a Cohen's kappa of 0.79 (95% CI 0.62–0.95; p <0.001). For intra-reader variability, a second round of readings was performed by the second reader after a 9-month interval, demonstrating substantial to almost perfect reliability (κ=0.83, 95% CI 0.76–0.88; p<0.001). Data collection For the study, the electronic medical records were accessed as described in Appendix B. The first presentation at the Department for Pneumology for evaluation of (suspected) sarcoidosis was considered the baseline date. End-point evaluation The primary end point of our study was disease progression. For the primary analysis, disease progression was concluded if at least one of the following three criteria was fulfilled at the corresponding follow-up visit at months 3, 6, 12, 18, 24 and 36: a.Therapy escalation, as either an increase in corticosteroid dose or the addition of a new immunosuppressive drugb.Decline of forced expiratory volume in 1 s (FEV1) of > 10% compared with the previous best documented visit or > 5% absolute compared with the previous documented visit, orc.Decline of diffusing capacity of the lung for carbon monoxide (DLCO) of > 20% compared with the previous best documented visit or > 15% absolute compared with the previous documented visit.For a sensitivity analysis, we included two other features in addition to these three parameters: Clinical judgement by the treating physician, orIncrease in soluble interleukin-2 receptor (sIL-2R) and/or neopterin of > 25% compared with the previous documented visit.Disease progression was concluded if at least one of the three original criteria or one of the two newly added criteria was fulfilled. a.Therapy escalation, as either an increase in corticosteroid dose or the addition of a new immunosuppressive drug Therapy escalation, as either an increase in corticosteroid dose or the addition of a new immunosuppressive drug b.Decline of forced expiratory volume in 1 s (FEV1) of > 10% compared with the previous best documented visit or > 5% absolute compared with the previous documented visit, or Decline of forced expiratory volume in 1 s (FEV1) of > 10% compared with the previous best documented visit or > 5% absolute compared with the previous documented visit, or c.Decline of diffusing capacity of the lung for carbon monoxide (DLCO) of > 20% compared with the previous best documented visit or > 15% absolute compared with the previous documented visit. Decline of diffusing capacity of the lung for carbon monoxide (DLCO) of > 20% compared with the previous best documented visit or > 15% absolute compared with the previous documented visit. Clinical judgement by the treating physician, or Clinical judgement by the treating physician, or Increase in soluble interleukin-2 receptor (sIL-2R) and/or neopterin of > 25% compared with the previous documented visit. Increase in soluble interleukin-2 receptor (sIL-2R) and/or neopterin of > 25% compared with the previous documented visit. Statistical analysis Continuous variables are presented as mean±standard deviation (sd) and/or median and interquartile range (IQR) as appropriate. Categorical variables are reported as frequencies and percentages. Group comparisons at baseline were performed using the Chi-squared test for categorical variables and the Kruskal–Wallis test for continuous variables. Censored data were analysed using Cox regression. Further details are provided in Appendix C. Results The final cohort consisted of 232 patients (mean age 50.1±14.9 years, 45.7% male), each patient having had one CT scan (table 1). The median interval between CT and lung function testing was 23 days (IQR 7–58 days). The presence of intravenous contrast (non-contrast (n=66, 28.4%) versus contrast-enhanced (n=166, 71.6%)) had no effect on the detection of ground glass opacities (GGO) (non-contrast: 16.7% present (11/66), contrast-enhanced: 15.7% present (26/166); p=0.84). The mean slice thickness was 2 mm. For analysis, slice thickness was dichotomised into ≤ 1.5 mm (n=144) and > 1.5 mm (n=88) in accordance with recommendations for HRCT. There was no significant impact of slice thickness on GGO detection (≤ 1–5 mm: 16.7% present (24/144), > 1.5 mm: 14.8% present (13/88); p=0.85). Based on the CT scans, the patients were broadly classified into two primary subgroups: non-fibrotic (n=211, 91.0%) and fibrotic (n=13, 5.6%) (tables 1 and 2). A small proportion of patients could not be confidently categorised into either of these two groups and were categorised as unclear (n=8, 3.4%). The patients’ demographics were comparable between groups. There was no difference in involvement of other organs between the phenotypes (table S1). TABLE 1Baseline characteristics of patientsNon-fibrotic (N=211)Unclear (N=8)Fibrotic (N=13)Total (N=232)p-value Age at first diagnosis, years Mean±sd50.1±14.947.8±11.552.8±16.750.1±14.90.88 Gender, n (%) Male100 (47.4)0 (0.0)6 (46.2)106 (45.7)0.03Female111 (52.6)8 (100.0)7 (53.8)126 (54.3) Smoking status, n (%) No84 (39.3)4 (50.0)4 (30.8)91 (39.2)0.19Active smoker30 (14.2)0 (0.0)0 (0.0)30 (12.9)Former smoker63 (29.9)1 (12.5)7 (53.8)71 (30.6)Missing35 (16.6)3 (37.5)2 (15.4)40 (17.2) Pack-years Mean±sd6.9±12.51.8±4.09.2±9.26.9±12.20.19 BMI at first visit, kg·m−2 Mean±sd28.1±5.927.1±3.826.3±4.828.0±5.80.57 FEV1/FVC, % Mean±sd75.9±7.574.8±6.667.2±10.675.4±7.90.005Median (IQR, range)76.7 (72.1–80.6, 45.6–92.4)71.50 (69.7–81.9, 68.4–84.2)65.8 (62.0–71.3, 48.2–87.4)76.5 (71.1–80.5, 45.6–92.4) FEV1, % Mean±sd88.4±20.283.3±13.364.1±18.286.8±20.6<0.001Median (IQR, range)88.2 (74.8–104.0, 37.5–155.0)82.3 (72.4–91.3, 67.5–106.8)63.5 (48.9–79.1, 37.2–95.4)85.9 (73.6–102.8, 37.2–155.0) FVC, % Mean±sd95.1±19.687.6±15.479.8±21.393.9±19.80.016Median (IQR, range)97.0 (81.8–108.8, 34.3–152.0)83.1 (74.7–102.5, 69.9–110.2)84.5 (74.1–91.9, 36.5–119.4)95.4 (81.3–108.5, 34.3–152.0) TLC, % Mean±sd87.5±13.776.3±11.675.40±15.086.4±14.00.002Median (IQR, range)87.4 (79.5–96.0, 42.8–127.6)78.2 (67.1–85.2, 58.0–91.2)72.6 (68.0–87.4, 39.3–95.3)86.8 (77.7–95.2, 39.3–127.6) DLCOcSB, % Mean±sd80.73±14.4476.5±7.266.2±19.179.8±14.90.038Median (IQR, range)80.4 (72.0–89.8, 38.5–131.4)79.1 (70.7–81.9, 64.8–83.4)65.8 (54.9–82.2, 24.9–89.7)80.2 (71.3–88.9, 24.9–131.4) sIL-2R, U·mL−1 Mean±sd1138.3±1022.1677.4±267.81270.9±810.71126.1±997.30.284Median (IQR, range)852 (524–1314, 8–7475)641 (450–844, 376–1174)1424 (597–2011, 286–2380)845 (523–1325, 8–7475) Neopterin, nmol·L−1 Mean±sd18.5±22.714.1±6.516.2±7.618.2±21.70.84Median (IQR, range)12.8 (9.1–19.8, 2.0–242.0)13.2 (9.6–17.8, 6.0–25.3)14.0 (10.5–23.3, 8.5–29.7)12.8 (9.1–19.8, 2.0–242.0) CT slice thickness, mm Mean±sd2.04±1.381.47±0.951.73±0.972.00±1.350.26Median (IQR, range)1.3 (1.0–3.0, 0.8–5.0)1.0 (1.0–2.0, 0.8–3.0)1.0 (1.0–3.0, 1.0–3.0)1.3 (1.0–3.0, 0.8–5.0)BMI: body mass index; FEV1: forced expiratory volume in 1 s; FVC: forced vital capacity; IQR: interquartile range; TLC: total lung capacity; DLCOcSB: diffusing capacity of the lung for carbon monoxide measured by the single breath method; sIL-2R: soluble interleukin-2 receptor; CT: computed tomography. p-values represent the values of the Chi-squared test for categorical data and the Kruskal–Wallis test for continuous variables. Baseline characteristics of patients BMI: body mass index; FEV1: forced expiratory volume in 1 s; FVC: forced vital capacity; IQR: interquartile range; TLC: total lung capacity; DLCOcSB: diffusing capacity of the lung for carbon monoxide measured by the single breath method; sIL-2R: soluble interleukin-2 receptor; CT: computed tomography. p-values represent the values of the Chi-squared test for categorical data and the Kruskal–Wallis test for continuous variables. TABLE 2Lung function by phenotypeLung functionCT-phenotypeNormalRestrictiveObstructiveMixedNormal/diffusion only (DLCO missing)Diffusion onlyTotal Non-fibrotic Normal21 (42.0)9 (18.0)4 (8.0)4 (8.0)5 (10.0)7 (14.0)50 Scattered nodules15 (31.3)13 (27.1)9 (18.6)1 (2.1)3 (6.25)7 (14.6)48 Peribronchovascular micronodules13 (29.5)12 (27.3)7 (15.9)1 (2.3)3 (6.8)8 (18.2)44 Perifissural nodules10 (45.8)2 (8.7)2 (8.7)1 (4.3)3 (13.0)5 (21.7)23 Random nodules4 (26.7)3 (20.0)1 (6.7)0 (0.0)1 (6.7)6 (40.0)15 Larger peribronchovascular nodules2 (20.0)1 (10.0)1 (10.0)0 (0.0)1 (10.0)5 (50.0)10 Predominant consolidation4 (44.4)0 (0.0)4 (44.4)1 (11.1)0 (0.0)0 (0.0)9 Diffuse GGO1 (16.7)2 (33.3)1 (16.7)1 (16.7)0 (0.0)1 (16.7)6 Others1 (16.7)2 (33.3)1 (16.7)0 (0.0)0 (0.0)2 (33.3)6 Unclear Bronchocentric reticulation0 (0.0)1 (33.3)0 (0.00)1 (33.3)1 (33.3)1 (33.3)3 Predominant consolidation2 (66.7)0 (0.0)0 (0.0)1 (33.3)0 (0.0)0 (0.0)3 Peribronchovascular micronodules0 (0.0)1 (50.0)0 (0.0)0 (0.00)0 (0.0)1 (50.0)2 Fibrotic PMF-like0 (0.0)0 (0.0)2 (28.7)5 (71.4)0 (0.0)0 (0.0)7 Bronchocentric reticulation1 (25.0)0 (0.0)1 (25.0)1 (25.0)0 (0.0)1 (25.0)4 Others1 (50.0)1 (50.0)0 (0.0)0 (0.0)0 (0.0)0 (0.0)2Data are presented as n (%). CT: computed tomography; DLCO: diffusing capacity of the lung for carbon monoxide; GGO: ground glass opacities; PMF: progressive massive fibrosis. Lung function by phenotype Data are presented as n (%). CT: computed tomography; DLCO: diffusing capacity of the lung for carbon monoxide; GGO: ground glass opacities; PMF: progressive massive fibrosis. Radiological phenotypes Non-fibrotic phenotypes Among those patients with the non-fibrotic phenotype (n=211, 91.0%), 50 (21.5%) exhibited no visible parenchymal involvement on CT. Nodular patterns were the most prevalent imaging phenotype observed within the cohort (n=142, 61.2%). Scattered nodules were the most common phenotype, identified in 48 patients (20.7%), none of whom exhibited fibrotic changes. This phenotype showed discrete nodules, typically ranging from 3 mm to 3 cm. Generally, this phenotype exhibited fewer than 15 nodules scattered throughout the pulmonary parenchyma, although a subset of patients (n=13) had more than 15 nodules without any lobar predilection. Peribronchovascular micronodules were the second most common subtype, identified in 46 patients (19.8%). While most of these patients did not exhibit fibrotic changes, subtle bronchocentric reticulations were observed around the nodular lesions in 2 of 46 patients (4.3%), who were classified as unclear for fibrosis. The nodules were typically smaller than 3 mm, and most patients showed more than 15 nodules. Larger nodules were observed in 10 patients (4.3%). In most patients, the perilymphatic nodules had coalesced to form larger nodular opacities in a peribronchovascular distribution. No compelling signs of fibrosis were noted in this group (figure 1). Computed tomography-based phenotypes in sarcoidosis as per the recent position paper. a) Peribronchovascular micronodules: high-resolution computed tomography (HRCT) showing numerous micronodules in a typical perilymphatic distribution. b) Larger nodules: HRCT showing typical larger nodules in a patient. Also notable are large mediastinal lymph nodes. c) Scattered nodules: HRCT showing a scattered nodule (blue arrow). d) Predominant consolidation: diffuse nodules seen in the left lung along with consolidation with a positive air bronchogram in the right upper lobe. e) Bronchocentric reticulations: predominant bihilar architectural distortion with dilatation of airways, which show airway distortion. f) Progressive massive fibrosis pattern: dense bronchocentric masses bilaterally with associated architectural distortion and posterior displacement of the hili. Predominant consolidations were present in 12 patients (5.2%). These consolidations were primarily peribronchovascular and perihilar, formed by the coalescence of nodules. While nine patients in this group did not show clear signs of fibrosis, three patients exhibited airway distortion and were categorised as unclear for fibrosis. Additionally, two patients with predominant subpleural consolidations resembling an organising pneumonia pattern were included in this category. In addition to the above-described non-fibrotic phenotypes described as per the Delphi consensus statement, we noted the following phenotypes. Isolated perifissural nodules (PFNs) were observed in 23 patients (9.9%). This phenotype included patients with a few or multiple PFNs or beaded fissures without any other recognisable pattern. This finding may represent enlarged intrapulmonary lymph nodes. All patients were non-fibrotic. Randomly distributed nodules (miliary pattern) were detected in 15 patients (6.5%). This phenotype consisted of multiple micronodules (< 3 mm) dispersed throughout the pulmonary parenchyma, without a clear perilymphatic predilection. Diffuse GGO were the dominant pattern observed in six patients (2.6%), all of whom were classified as non-fibrotic (figure 2). Other computed tomography (CT)-based phenotypes noted in our cohort. a) Perifissural nodules: axial high-resolution computed tomography (HRCT) section showing predominant subpleural nodules (blue arrow). b) Diffuse ground-glass opacities: CT scan showing a diffuse, hazy increase in the lung density, presumably due to diffusely scattered fine nodules throughout the lung parenchyma. c) Random nodules: axial HRCT image showing multiple micronodules without any predilection for centrilobular or perilymphatic pattern. Fibrotic phenotypes Among the patients exhibiting a fibrotic phenotype (n=13), two major patterns were identified. Progressive massive fibrosis (PMF)-like pattern was observed in seven patients (3.0%), all of whom were classified as fibrotic. This phenotype was characterised by bilateral, dense bronchocentric masses or consolidations, accompanied by signs of parenchymal distortion and fibrosis, including airway distortion and cicatricial emphysema. These findings demonstrated a predilection for the upper lobes in four patients. Bronchocentric reticulation was identified in seven patients (3.0%), of whom four were classified as fibrotic and three were classified as unclear for fibrosis because the reticulations were subtle without other signs of fibrosis or could not be differentiated from septal thickening. This phenotype was characterised by bronchocentric reticulations, which may or may not be accompanied by peribronchovascular nodular changes. Lower zone-predominant usual interstitial pneumonia (UIP)-like fibrosis with honeycombing was seen in one patient, while another patient showed probable UIP pattern of fibrosis with perifissural noduli. Minor apical capping with pleuroparenchymal fibroelastosis-like changes were noted in only five patients who had no predilection for any particular phenotype (both fibrotic and non-fibrotic). End points The median follow-up time for the entire cohort was 24 months (IQR 6–36 months). In the non-fibrotic group, the median follow-up was 24 months (IQR 6–36 months), while fibrotic group had a longer median follow-up of 36 months (IQR 18–36 months). A total of 160 patients (69.0%) fulfilled at least one progression criterion used for the primary analysis until follow-up at month 36, while a total of 76 (32.8%) patients showed more than one criterion of progression during follow-up. The median time to progression was 12 months. Using the broader definition (sensitivity analysis of the primary end point), a total of 176 patients (75.9%) showed progression. 141 (60.8%) patients showed more than one criterion of progression during follow-up. The median time to progression was 6 months. Disease progression in different radiological phenotypes To assess disease progression, we first compared patients with initial fibrotic versus non-fibrotic radiological phenotypes. All patients (100%, 95% CI 75–100%) with the fibrotic phenotype demonstrated progression in at least one criterion at 36 months. By contrast, those with the non-fibrotic phenotype exhibited a more gradual increase, with progression of 67% (95% CI 60–73%) at 36 months (figure 3a, figure S1). Primary end point: progression among computed tomography groups over a follow-up period of 36 months with progress in at least one criterion (therapy escalation, worsening forced expiratory volume in 1 s and decline of diffusing capacity of the lung for carbon monoxide). a) Non-fibrotic (n=211), unclear (n=8) and fibrotic (n=13). b) Normal (n=50), nodular (n=142) and non-nodular (n=40). In a second approach, we focused on nodular and non-nodular radiological phenotypes. Peribronchovascular micronodules, scattered nodules, random nodules, isolated perifissural nodules and larger nodules were included in the nodular group (n=142). The remaining phenotypes except normal were included in the non-nodular group (n=40). Patients with the non-nodular group exhibited the highest and most rapid rate of progression over time, reaching 88% (95% CI 73–96%) at 36 months. In contrast, the nodular phenotype showed a lower and more gradual progression, 64% (95% CI 56–72%) at 36 months. The phenotype group “nodular” had similar progression to patients with normal CT (68%, 95% CI 53–80%) at 36 months; (figure 3b, figure S2). These findings were similar in the sensitivity analysis with the extended definition of progression (figure 4, figures S3 and S4). Particularly, it confirmed that the phenotype group “nodular” had the same risk for progression as patients with normal CT (74% versus 72% at 36 months). Sensitivity analysis: progression among computed tomography groups over a follow-up period of 36 months with progress in at least one criterion (therapy escalation, worsening forced expiratory volume in 1 s, decline of diffusing capacity of the lung for carbon monoxide, clinical judgement and increase of soluble interleukin-2 receptor and/or neopterin). a) Non-fibrotic (n=211), unclear (n=8) and fibrotic (n=13). b) Normal (n=50), nodular (n=142) and non-nodular (n=40). A small cohort size of some of the individual Delphi phenotypes precluded a detailed analysis of progress in each of these separately. Figure S5 depicts the progression pattern in each of the CT phenotypes, mirroring the overall impression that fibrotic phenotypes performed worse. Prediction of disease progression In an unadjusted Cox regression model, the presence of fibrotic disease was associated with a more than twofold increased risk of progression compared with non-fibrotic disease (hazard ratio (HR) 2.07, 95% CI 1.17–3.67; p=0.04). In contrast, age, gender, body mass index and lung function at baseline were not associated with progression (all p≥0.24). After adjustment for potential confounder, the presence of fibrosis remained associated with disease progression (HR 2.31; p=0.04). Addition sub-analyses are provided in Appendix D, Tables S2–4. Discussion A recent multinational Delphi consensus study identified seven distinct CT phenotypes in sarcoidosis as a potential framework for future research [4]. In our study, we applied these morphological phenotypes in a tertiary referral cohort and reported four key findings: 1) we described the prevalence of individual CT phenotypes in a well-characterised cohort of more than 200 sarcoidosis patients and identified additional patterns that did not reach a consensus in the Delphi exercise; 2) grouping radiological phenotypes into non-fibrotic and fibrotic groups emerged as the most important step pertaining to disease behaviour; 3) nodular patterns showed a similar disease behaviour to radiologically normal lungs with respect to disease progression over a follow-up period of 36 months; and 4) inter-reader variability was moderate for individual phenotypes but improved to good agreement when assessing fibrotic versus non-fibrotic categories. Attempts to derive clinically meaningful imaging phenotypes of sarcoidosis and thereby predict disease course are not new. Early efforts centred on radiographic classification, exemplified by the landmark Scadding system, which is still widely used [12]. However, Scadding stage correlates poorly with patient-reported disease severity, lung function tests and the need for patient treatment [11, 13], and it also has limited interobserver reproducibility: these factors have driven increasing interest in CT-based classifications [10, 14]. Indeed, discordance between chest radiograph and HRCT staging has been reported in 40–50% of cases [15, 16], and HRCT stage, not radiographic stage, correlated with DLCO. CT patterns in sarcoidosis have also been classified into typical and atypical features, primarily to facilitate timely diagnosis and to familiarise radiologists with the broad spectrum of disease manifestations [17–21]. Typical CT patterns of sarcoidosis are characterised by peribronchovascular micronodules, with or without interlobular septal thickening, whereas atypical patterns include larger nodules and perihilar masses, which are more often associated with fibrosis. In addition, certain findings have been described as “active” patterns – such as nodules, GGO and alveolar consolidations – to distinguish them from features of irreversible fibrosis, including honeycomb-like cysts, bullae, broad septal bands, architectural distortion, volume loss and traction bronchiectasis [17]. Another study describing HRCT findings in patients with end-stage sarcoidosis identified three distinct fibrotic patterns associated with specific lung function profiles: 1) central bronchial distortion, 2) honeycombing, and 3) diffuse linear opacities. Notably, the Delphi phenotype of “bronchocentric reticulation” appears to encompass elements of both groups 1 and 3, effectively merging them into a single category [23, 24]. The Delphi consensus approach used to define HRCT phenotypes in sarcoidosis has several advantages. It provides a structured framework to capture expert knowledge in areas where empirical evidence is limited, especially for a disease such as sarcoidosis. It also ensures multinational representation; this is important in the context of sarcoidosis, which shows different prevalence in different ethnicities. Such an exercise also enables terminology to be standardised for future studies. This is particularly valuable in sarcoidosis, where imaging manifestations are highly heterogeneous and lack a unifying classification system. In this context, the main strength of the Delphi exercise lies in identifying and proposing various phenotypes that could be broadly grouped into non-fibrotic or fibrotic. However, Delphi-based methods also have inherent drawbacks; for example, they rely on experts’ judgement rather than being data driven. A further limitation of the Delphi consensus is the composition of its expert panel, the members of which were predominantly from Europe and North America, potentially limiting global representativeness. In addition, the framing of survey questions might have introduced bias by constraining the range of phenotypes considered, potentially resulting in excessive granularity and blurring of some categories. The resulting classification into discrete phenotypes remains to some extent artificial. Whether the distinct radiological phenotypes represent distinct disease entities or snapshots of different stages of disease evolution requires further investigation through longitudinal CT studies. In clinical practice, some cases demonstrate overlapping imaging features, making a strict separation of phenotypes challenging. In our experience, some patterns such as larger nodules, consolidation and PMF proved especially difficult to classify. This inherent overlap likely contributes to the only moderate inter-reader agreement observed and highlights the limitations of consensus-based classifications, suggesting that clearer definitions or broader categories may be needed to improve reproducibility in sarcoidosis. Similar results of inter-reader agreement were shown in a preliminary analysis from an international multicentre study in 203 patients, with a kappa of 0.433 across four readers for individual phenotypes, which improved to 0.55 while grouping the nodular phenotypes [25]. A recent Dutch study evaluating inter-reader reliability for scoring individual CT phenotypes in 45 sarcoidosis patients reported substantial agreement (κ=0.69, 95% CI 0.62–0.77; p<0.001), which further improved when patients were dichotomised into fibrotic versus non-fibrotic groups (κ=0.78, 95% CI 0.61–0.94; p<0.001). The scans were preselected to avoid ambiguous cases, which may partly explain the higher kappa values reported [26]. Some of the phenotypes that did not receive a majority consensus agreement, such as GGO and miliary pattern, emerged as a separate phenotype in our study. In our cohort, a more robust classification emerged when patients were grouped into fibrosis versus non-fibrosis or nodular versus non-nodular phenotypes. These simplified categories not only showed higher interobserver agreement but also correlated more closely with clinical outcomes. Considering the heterogeneous nature of sarcoidosis, and in order to represent various important areas of disease progression, we used different domains to assess sarcoidosis disease progression, including the need for therapy escalation, lung function parameters, clinical deterioration as assessed by the treating physician and serological parameters. Currently, no consensus definition exists for disease progression in sarcoidosis. Previous studies have attempted to identify clinical and radiological markers of progression in sarcoidosis [27, 28]. One study proposed a scoring system based on four CT features – nodules, interlobular septal thickening, consolidation and GGO – to correlate with disease activity [22]. Another study developed a staging system for prognostication by integrating the Composite Physiologic Index, a weighted index of pulmonary function variables, with two HRCT parameters: fibrosis and the ratio of main pulmonary artery to ascending aorta diameter [27]. In this context, our study is noteworthy because it assesses disease progression in different domains, including but not limited to lung function parameters. Fibrotic sarcoidosis predisposes to a more rapid and more likely progressive disease with high mortality [29]. Accordingly, in our cohort, the initial presentation with fibrotic alterations on CT was associated with a more rapid disease progression on all assessed domains. While different nodular patterns in sarcoidosis may be important for considering appropriate differential diagnosis, their differentiation appears to be of secondary importance concerning disease behaviour. In our study, we did not censor patients who were lost to follow-up, given the low mortality rate in sarcoidosis of approximately 1–8% in referral settings and even lower in peripheral settings [30–33]. Thus, no more than 18 of our patients might have died during the follow-up period and would have been misclassified using our approach of not censoring. While this introduces a potential misclassification bias, the random nature of the drop-out would suggest a minimal impact on the overall conclusion. Furthermore, higher mortality rates have been shown among patients who had fibrotic sarcoidosis previously [34]. Among our population, a higher median follow-up rate was seen in this group than in the non-fibrotic group, thus reducing the risk of patients being lost as a result of death. Furthermore, this longer follow-up in the fibrotic group further supports our assumption that patients who did not return to our tertiary care centre felt no worsening symptoms. Thus, we feel that the risk of misclassification is relatively low and supports our analytical approach. Our findings should be interpreted in the context of several limitations, as detailed in Appendix E. To our knowledge, ours is one of the first studies describing the CT phenotypes in sarcoidosis by incorporating the recent consensus study and linking it to clinically relevant events. It therefore allows a better understanding of the clinical strengths and features of the proposed radiological phenotypes and may pave the way for further studies investigating the interplay between radiological, lung functional and clinical phenotypes in sarcoidosis and their impact on disease behaviour. In conclusion, CT-based classification of sarcoidosis based on the recent recommendations seems feasible, although some additional phenotypes and overlaps between categories were observed, underscoring the importance of further refinement of this classification. Splitting sarcoidosis into various CT-based phenotypes serves as an initial step for future research in defining various aspects of pulmonary sarcoidosis. However, lumping these phenotypes into clinically meaningful groups such as fibrotic versus non-fibrotic and nodular versus non-nodular patterns could be the next logical step to understand disease behaviour.