CT-based phenotyping of COVID-19: cluster analysis of pulmonary and extrapulmonary imaging markers from a multicentre retrospective cohort study

Abstract

Coronavirus disease 2019 (COVID-19) displays a highly variable clinical course despite advancements in vaccination and antiviral therapies. Chest computed tomography (CT) has become a valuable tool for diagnosing and predicting COVID-19 severity; however, limited studies have explored integrating pulmonary and extrapulmonary CT markers to identify distinct clinical phenotypes. In this study, we aimed to evaluate the utility of cluster analysis, using quantitative pulmonary and extrapulmonary CT indicators, to classify patients with COVID-19 into distinct phenotypic clusters and assess their clinical relevance in predicting disease severity and outcomes. The primary outcome was the rate of critical outcomes (requiring high-flow oxygen therapy or invasive ventilator support or death). In this multicentre, retrospective cohort study, we analysed 1,034 patients with COVID-19 from four hospitals in Japan. Hierarchical cluster analysis was performed using age, sex, and seven imaging markers: pneumonia volume, muscle area, muscle density, subcutaneous and visceral fat indices, bone density, and coronary artery calcification score. Clinical characteristics, laboratory findings, and outcomes were compared across the identified clusters. Four distinct clusters were identified. Cluster 1 consisted of younger individuals with minimal pneumonia and favourable extrapulmonary organ markers, exhibiting the best clinical outcomes. Cluster 2 included younger patients with high pneumonia volume and visceral fat accumulation, exhibiting poor respiratory outcomes. Cluster 3 comprised older individuals with mild fat accumulation and low bone density, with intermediate severity. Cluster 4 presented the highest pneumonia volume, extensive visceral fat, and coronary artery calcification, resulting in the worst overall prognosis, including the highest mortality and in-hospital complications. Clustering based on pulmonary and extrapulmonary CT indicators enabled the precise classification of patients with COVID-19 into clinically significant subgroups with distinct outcomes. This study highlights the importance of integrating multiple imaging markers for disease phenotyping and prognosis.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-26187-7.

Introduction

Coronavirus disease 2019 (COVID-19) was first identified in Wuhan, China, in late 2019 and rapidly became a global pandemic in 2020¹. The severity of the disease has decreased due to viral strain evolution and advancements in vaccination²; however, its clinical course remains highly variable^3,4. Given the recurring waves of infection and the substantial number of affected individuals, numerous studies have explored various clinical characteristics and biomarkers associated with disease severity^5–7. Chest computed tomography (CT) was initially reported as a valuable tool for diagnosing and prognosticating COVID-19⁸. Furthermore, qualitative and quantitative assessment of pneumonia extent using CT is valuable for predicting disease severity^9,10, leading to its widespread adoption in clinical practice. Recent advancements in chest CT analysis techniques demonstrate its utility in evaluating extrapulmonary organ involvement in various lung diseases, including chronic obstructive pulmonary disease (COPD) and bronchial asthma^11,12. In addition, several studies on COVID-19, including ours, have reported associations between clinical outcomes and extrapulmonary organ indicators such as muscle area, muscle CT density, visceral fat (VAT) area, bone CT density, and coronary artery calcification^13–17. Cluster analysis, which seeks to organize information to classsify heterogeneous groups of variables into relatively homogeneous groups, has been proposed to examine phenotypic heterogeneity in various diseases^18,19. Due to the diversity of the clinical course of COVID-19, the utility of clustering using clinical information has been reported²⁰. Recently, clustering using CT indicators has also been performed for various diseases. For instance, studies on COPD and interstitial pneumonia have used chest CT imaging features to classify patients into distinct clusters, revealing differences in prognosis and treatment responses^21,22. However, few studies have conducted cluster analyses using CT indicators in patients with COVID-19^23,24, and no studies have incorporated extrapulmonary organ indicators. Therefore, we hypothesise that clustering pulmonary and extrapulmonary organ indicators could identify distinct clinical phenotypes and serve as a valuable tool for predicting clinical outcomes in patients with COVID-19.

In this study, we aimed to evaluate the utility of clustering based on previously reported CT prognostic indicators, including pulmonary indicators (pneumonia volume) and extrapulmonary organ indicators (erector spinae muscle [ESM] index, ESM mean, subcutaneous adipose tissue [SAT] index, epicardial adipose tissue [EAT] index, bone mineral density [BMD], and Agatston score).

Methods

Study design and setting

The detailed protocol of this study has been reported previously²⁴. This multicentre retrospective study performed a secondary analysis of the data of all COVID-19 cases obtained from the Japan COVID-19 Task Force. All participants in this study provided informed consent. The Japan COVID-19 Task Force collected clinical information on patients with COVID-19 aged > 18 years diagnosed by polymerase chain reaction or antigen testing from four hospitals across Japan. The study flow is shown in Additional file 1. Of the 1410 patients identified, 376 were excluded due to inappropriate images for analysis or a lack of CT imaging or height data; hence, 1034 patients were included in the analysis.

The primary outcome was the critical outcome, while the secondary outcomes included oxygen demand, mechanical ventilation, and mortality. Critical outcomes were defined as the use of invasive positive pressure ventilation (IPPV); extracorporeal membrane oxygenation; high-flow oxygen, including high-flow nasal canal, or non-invasive positive pressure ventilation; or patient mortality after hospitalization to assess clinically important outcomes in patients with COVID-19²⁵.

Data collection

Details regarding data collection have been reported in our previous study²⁴. Here, we provide a brief summary: The following information was extracted from the electronic case record form: age, sex, height, weight, clinical signs and symptoms, laboratory findings on admission, comorbidities, and treatment details. Patient outcomes were monitored throughout the hospitalization period. All laboratory tests were performed according to the clinical care needs of the patients. Symptoms, including upper and lower respiratory tract symptoms and signs, were considered at the time of referral and admission and throughout the hospitalization period. Upper respiratory symptoms included sore throat, nasal discharge, dysosmia, and dysgeusia; lower respiratory symptoms included cough, sputum production, and dyspnoea. Laboratory results were collected within 48 h of the initial visit or admission. The attending physicians at each facility reviewed and evaluated the chest CT images for qualitative pneumonia. A team of respiratory clinicians reviewed the data and contacted clinicians for missing core data. Missing or absent data on patient background were recorded as unknown.

CT acquisition

The CT acquisition methods have been published elsewhere¹². In summary, all CT images were obtained after full inspiration. Images of the entire lung with a slice thickness of 1–5 mm were reconstructed using standard kernels. The CT scanners used were SOMATOM series (Siemens Healthineers), Aquilion series (Canon Medical Systems), Revolution series (GE Healthcare), Discovery series (GE Healthcare), and BrightSpeed (GE Healthcare).

Image analysis

The procedures for image analysis have been published in our prior study²⁶. Pneumonia and total lung segmentation were performed using SYNAPSE VINCENT software (FUJIFILM, Tokyo, Japan). ESM, SAT, EAT, BMD, and CAC were quantitatively evaluated using ImageJ (Fiji) software for the manual masking of target regions and custom-made Python scripts for automatic area and density calculation. Representative images are shown in Fig. 1, and details are described in our previous studies^{13–16,26,27}.

Fig. 1.

Representative CT images for measuring (a) pneumonia volume, (b) subcutaneous adipose tissue, (c) coronary artery calcification, (d) epicardial adipose tissue, (e) bone mineral density, and (f) erector spinae muscles in patients with COVID-19.

Cluster analysis

We selected clinically relevant imaging markers reportedly associated with the severity or prognosis of COVID-19, namely, pneumonia volume, ESM, EAT, and SAT indices, ESM mean, BMD, and Agatston score^13–16,27. Hierarchical cluster analysis, using age, sex, and the seven imaging markers mentioned above, was performed using Ward’s minimum-variance method²⁸. The results are graphically presented using a dendrogram²⁹.

Statistical analysis

Statistical analyses were conducted as described in our prior work^12–14. Data are presented as means ± standard deviation (SD). Data were compared among groups using analysis of variance and χ² tests. Statistical significance was set at p < 0.05. All data were analysed using JMP 17 software (SAS Institute, Cary, NC, USA).

Results

Baseline patient characteristics

Table 1 presents the baseline characteristics of the enrolled patients. The mean age was 55.3 years, and 69.2% of the patients were male. The prevalence of lifestyle-related comorbidities such as hypertension, diabetes mellitus, and hyperuricemia was 31.2%, 19.3%, and 10.7%, respectively. The pneumonia volume within 48 h of hospitalisation was 14.8 ± 16.3%, and the ESM index, ESM mean, SAT index, EAT index, BMD, and Agatston score were 1269.8 ± 380.6 cm²/m², 38.1 ± 12.7 HU, 15.6 ± 9.4 cm²/m², 5.3 ± 3.5 cm²/m², 177.2 ± 53.3 HU, and 676.4 ± 2757.6 HU, respectively.

Table 1.

Demographic characteristics.

	All (N = 1034)
Age	55.3 ± 16.4
Sex (Male) ,%	69.2
BMI (kg/m2)	25.0 ± 6.2
Smoker ,%	46.3
Hypertension ,%	31.2
Diabetes ,%	19.3
Cardiovascular disease ,%	9.4
Malignancy ,%	9.5
COPD ,%	2.8
Asthma ,%	7.8
Hyperuricemia ,%	10.7
Chronic liver disease ,%	2.8
Chronic kidney disease ,%	7.9
Pneumonia volume (%)	14.8 ± 16.3
ESM index (cm2/m2)	1269.8 ± 380.6
ESM mean (HU)	38.1 ± 12.7
SAT index (cm2/m2)	15.6 ± 9.4
EAT index (cm2/m2)	5.3 ± 3.5
Bone mineral density (HU)	177.2 ± 53.3
Agatston score	676.4 ± 2757.6

Data are shown as mean ± standard deviation.

BMI, Body mass index; COPD, Chronic obstructive pulmonary disease; ESM, Erector spinae muscles; SAT, Subcutaneous adipose tissue; EAT, Epicardial adipose tissue.

Clustering based on pulmonary and extrapulmonary CT indicators

Figure 2 displays the dendrogram for clustering based on quantitative imaging markers of pulmonary and extrapulmonary organs. Based on visual inspiration, a threshold was arbitrarily determined to classify patients into four clusters. Figure 3 illustrates the comparison of CT indicators among these clusters, revealing significant differences among the four groups. Cluster 1 exhibited low pneumonia volume, high BMD, and low SAT/EAT index and Agatston score. Cluster 2 had a high pneumonia volume and a high SAT/EAT index. Cluster 3 presented low pneumonia volume, BMD, SAT/EAT index, and Agatston score. Cluster 4 had a high pneumonia volume, low SAT index and BMD, and high EAT index and Agatston score.

Fig. 2.

Dendrogram illustrating the results of the cluster analysis of 1034 patients with COVID-19 using Ward’s hierarchical clustering method.

Fig. 3.

Comparison of pulmonary and extrapulmonary CT indicators among the four clusters.

Comparison of clinical characteristics among clusters

Table 2 details the baseline clinical characteristics of each cluster. Cluster 1 was characterized by younger age, appropriate body mass index (BMI), and low comorbidity prevalence. Cluster 2 included younger patients with high BMI and high prevalence of hypertension, diabetes mellitus, cardiovascular disease, hyperuricemia, and chronic kidney disease. Cluster 3 consisted of old patients with comorbidity prevalence similar to that in Cluster 2. Cluster 4 had the oldest patients and the highest prevalence of the aforementioned comorbidities.

Table 2.

Comparison of baseline characteristics between clusters.

	Cluster 1 (N = 453)	Cluster 2 (N = 373)	Cluster 3 (N = 185)	Cluster 4 (N = 23)	P value
Age	46.4 ± 14.4	55.9 ± 12.5	74.0 ± 8.7	73.1 ± 11.5	< 0.0001
Sex (Male)	70.0	70.0	64.9	78.3	0.43
BMI (kg/m2)	23.1 ± 3.5	28.7 ± 7.9	22.1 ± 3.2	24.7 ± 5.2	< 0.0001
Smoker,%	42.4	49.5	48.4	54.6	0.16
Hypertension,%	13.7	42.7	47.6	59.1	< 0.0001
Diabetes,%	10.0	26.3	21.7	69.6	< 0.0001
Cardiovascular disease,%	4.4	9.7	15.4	56.5	< 0.0001
Malignancy,%	7.1	6.5	20.5	17.4	< 0.0001
COPD,%	1.6	1.9	7.7	4.4	0.0002
Asthma,%	7.1	8.5	7.8	8.7	0.90
Hyperuricemia,%	6.2	13.8	14.3	21.7	0.0004
Chronic liver disease,%	1.6	4.3	2.2	8.7	0.031
Chronic kidney disease,%	4.0	7.3	12.5	56.5	< 0.0001

Data are shown as mean ± standard deviation.

BMI, Body mass index; COPD, Chronic obstructive pulmonary disease.

A comparison of clinical symptoms among clusters revealed significant differences in upper and lower respiratory tract symptoms. Clusters 2 and 4 had few upper respiratory tract symptoms but a high frequency of lower respiratory tract symptoms (Additional file 2). Table 3 presents the comparison of blood test results. Significant differences were observed across several biomarkers associated with COVID-19 severity, including inflammatory markers, renal function, albumin (Alb), and brain natriuretic peptide (BNP)^29–32. Notably, Cluster 4 had the highest white blood cell count, neutrophil-to-lymphocyte ratio (NLR), creatinine (Cr), and BNP, while Alb the was lowest. These results suggest that clustering based on quantitative CT evaluation of pulmonary and extrapulmonary organs effectively identifies clinically distinct patient groups.

Table 3.

Comparison of laboratory values between clusters.

	Cluster 1 (N = 453)	Cluster 2 (N = 373)	Cluster 3 (N = 185)	Cluster 4 (N = 23)	P value
WBC (/µL)	5053.5 ± 1974.4	5883.4 ± 2896.9	5077.8 ± 1878.5	5439.1 ± 1981.0	< 0.0001
Neutrophil (/µL)	3485.6 ± 1821.5	4293.9 ± 2445.0	3614.5 ± 1849.1	4061.1 ± 1764.6	< 0.0001
Lymphocytes (/µL)	1116.3 ± 471.4	1064.2 ± 529.8	982.8 ± 448.4	849.3 ± 456.3	0.0023
AST (IU/L)	33.8 ± 33.5	47.3 ± 29.1	35.3 ± 22.8	47.2 ± 66.0	< 0.0001
ALT (IU/L)	31.8 ± 31.7	49.4 ± 37.9	26.5 ± 22.0	25.7 ± 40.1	< 0.0001
γ-GTP (IU/L)	53.9 ± 62.9	91.5 ± 94.5	44.5 ± 47.7	40.5 ± 34.1	< 0.0001
Alb (mg/dL)	3.9 ± 0.5	3.6 ± 0.6	3.6 ± 0.5	3.3 ± 0.5	< 0.0001
BUN (mg/dL)	14.5 ± 11.8	16.8 ± 12.2	19.2 ± 11.7	36.9 ± 21.3	< 0.0001
Cr (mg/dL)	1.1 ± 2.0	1.2 ± 1.9	1.2 ± 1.6	5.1 ± 5.3	< 0.0001
LDH (U/L)	233.7 ± 94.7	306.3 ± 123.7	243.4 ± 77.7	302.9 ± 120.4	< 0.0001
UA (mg/dL)	4.6 ± 1.4	5.1 ± 1.7	4.8 ± 1.8	6.1 ± 2.8	< 0.0001
CK (IU/L)	145.2 ± 679.2	199.0 ± 434.7	164.1 ± 267.8	635.7 ± 1525.6	0.0009
HbA1c (%)	5.9 ± 1.1	6.6 ± 1.3	6.2 ± 0.9	7.4 ± 1.6	< 0.0001
TG (mg/dL)	119.3 ± 65.6	145.1 ± 103.3	107.2 ± 61.2	137.8 ± 74.8	< 0.0001
PCT (ng/mL)	0.2 ± 1.4	0.3 ± 0.8	0.4 ± 1.7	1.5 ± 2.9	0.0006
D-dimer (µg/mL)	1.4 ± 3.5	1.8 ± 4.5	2.1 ± 5.5	1.7 ± 1.1	0.27
BNP (pg/mL)	17.5 ± 57.6	25.2 ± 71.9	63.6 ± 165.1	307.0 ± 546.2	< 0.0001
Ferritin (ng/mL)	461.9 ± 724.5	790.8 ± 954.2	455.1 ± 465.3	398.2 ± 421.0	< 0.0001
Fibrinogen (mg/dL)	442.4 ± 133.5	513.8 ± 144.8	470.8 ± 127.0	490.2 ± 143.8	< 0.0001
CRP (mg/dL)	2.9 ± 3.9	6.6 ± 7.0	4.8 ± 6.1	8.1 ± 8.3	< 0.0001

Data are shown as mean ± standard deviation.

WBC, White blood cell; AST, Aspartate aminotransferase; ALT, Alanine aminotransferase; BUN, Blood urea nitrogen; Cr, Creatinine; LDH, Lactate dehydrogenase; UA, Uric acid; CK, Creatine kinase; TG, Triglyceride; PCT, Procalcitonin; BNP, Brain natriuretic peptide; CRP, C-reactive protein.

Comparison of clinical outcomes among clusters

Figure 4 illustrates the comparison of clinical outcomes among the clusters. Significant differences were observed in critical outcomes, oxygen demand, mechanical ventilation, and mortality across the four groups. The proportion of critical outcomes was highest in the order of Cluster 4 > Cluster 2 > Cluster 3 > Cluster 1 (Fig. 4a). Among critical outcomes, oxygen demand (Fig. 4b) and ventilator use (Fig. 4c) were highest in Cluster 2, while mortality was highest in Cluster 4 (Fig. 4d). Figure 5 shows the comparison of in-hospital complications among the clusters. Significant differences were observed in bacterial infections, heart failure, thromboembolism, liver dysfunction, and renal dysfunction across the different clusters. Bacterial infections increased progressively from Cluster 1 to 4. Heart failure was most prevalent in Cluster 4. Thromboembolism was more common in Clusters 2, 3, and 4. Liver dysfunction was more frequent in Cluster 2. Renal dysfunction increased progressively from Cluster 1 to 4. These results suggest that Clusters 2 and 4 had the worst outcomes. Cluster 2 was associated with poorer respiratory outcomes, while Cluster 4 had higher rates of in-hospital complications and mortality.

Fig. 4.

Comparison of clinical outcomes among the four clusters. (a) Comparison of the rate of critical outcome. (b) Comparison of the rate of receiving supplementary oxygen. (c) Comparison of the rate of requiring mechanical ventilation. (d) Comparison of the mortality rate.

Fig. 5.

Comparison of the incidence of post-hospitalisation complications among the four clusters.

Discussion

This study is the first to perform cluster analysis using quantitative pulmonary and extrapulmonary CT indicators in patients with COVID-19. The analysis identified four distinct clusters based on imaging characteristics: Cluster 1 comprised younger patients with low pneumonia volume and minimal extrapulmonary organ complications on CT.

Cluster 2 included younger individuals with high levels of SAT and VAT and high pneumonia volume. Cluster 3 comprised older individuals with mild VAT accumulation, coronary artery calcification, and low pneumonia volume. Cluster 4 included older individuals with extensive VAT and coronary artery calcification and high pneumonia volume. Cluster 2 had worse respiratory-related outcomes. Cluster 4 had the highest risk of in-hospital multi-organ complications and mortality, indicating significant differences in disease severity and prognosis between the four clusters.

Cluster 1 exhibited the most favourable clinical outcomes, characterized by low pneumonia volume, high BMD, and low SAT index, EAT index, and Agatston score. These findings align with previous studies, including ours, which reported that these quantitative CT markers are associated with COVID-19 prognosis^13,15,16,27. Moreover, the patients in Cluster 1 were young and had a low prevalence of lifestyle-related diseases. This result corroborates prior research indicating that older age, hypertension, diabetes, and hyperuricemia are associated with worse prognosis in patients with COVID-19^33–36.

Cluster 2 had a similar age profile to Cluster 1; however the patients exhibited higher pneumonia volume and a high SAT/EAT index, leading to the second-worst clinical outcomes. Based on previous studies, including ours, visceral fat was associated with both the extent of pneumonia and disease severity^14,37; therefore, this distinction was primarily attributed to differences in fat accumulation, as identified on CT. In addition, Cluster 2 had a higher BMI and prevalence of hypertension, diabetes, and hyperuricemia compared to Cluster 1, reinforcing the robustness of this association. Notably, although Cluster 4 exhibited all the imaging markers associated with poor prognosis, Cluster 2 had comparable or even worse respiratory outcomes. In fact, several studies, including ours, have demonstrated that VAT accumulation correlates with pneumonia extent and prognosis^14,37. The association between VAT and pneumonia is supported by the following mechanisms: angiotensin-converting enzyme 2 receptors, which mediate severe acute respiratory syndrome coronavirus 2 entry into target cells, are highly expressed in the VAT³⁸. Furthermore, the VAT is a source of inflammatory cytokines, prolonging pulmonary inflammation^39,40. Cluster 2 had the highest oxygen demand and mechanical ventilation rate, supporting the correlation between fat accumulation and pneumonia severity.

Cluster 3 exhibited low pneumonia volume, BMD, SAT/EAT index, and Agatston score. Interestingly, despite low ESM index and BMD, both associated with COVID-19 severity^13,26, the pneumonia volume remained low, and clinical outcomes were better than those in Clusters 2 and 4. As the ESM index and BMD decline with ageing due to muscle degeneration and osteoporosis, Cluster 3 likely reflects age-related characteristics^41,42. However, while Clusters 1 and 3 had comparable pneumonia volume, Cluster 3 had worse clinical outcomes, likely due to age-related factors. The impact of ageing on COVID-19 severity has been attributed to age-related decline in immune cell function, leading to impaired immune responses⁴³ and endothelial dysfunction with aging, increasing viral entry into cells and exacerbating disease severity⁴⁴. Furthermore, renal dysfunction and cardiovascular events, including thromboembolic complications, are more frequent in older adults^45,46. In this study, Cluster 3 had a higher incidence of in-hospital complications than that in Cluster 1, which is consistent with these findings.

Cluster 4 exhibited high pneumonia volume, low ESM index, SAT index, and BMD, and high EAT index and Agatston score. This cluster exhibited the highest number of complications and the worst clinical outcomes. Cluster 4 displayed all imaging features previously reported to be associated with poor prognosis, correlating well with the observed clinical outcomes^{13–16,26,27}. Inflammatory markers, B-type natriuretic peptide, renal function markers, Alb, and respiratory symptoms have all been linked to COVID-19 prognosis^29–32,47. Cluster 4 exhibited significant abnormalities in these markers, supporting the hypothesis that CT-based clustering accurately captures disease severity. Notably, although no single CT indicator differed significantly between Clusters 2 and 4, Cluster 4 had the highest mortality and complication rates. This underscores the importance of evaluating multiple imaging markers collectively rather than relying on a single CT parameter.

Our findings highlight the clinical utility of clustering based on the quantitative analysis of pulmonary and extrapulmonary organs, enabling classification into distinct phenotypic subgroups with different prognoses. A key insight is the validation of the approach of integrating multiple imaging markers to predict COVID-19 severity. Previous studies have established the prognostic value of individual biomarkers, such as pneumonia extent, muscle markers, and fat markers^13–17. However, our study shows that pneumonia volume alone does not fully capture disease severity, as evidenced by the different clinical outcomes between Clusters 1 and 3 and Clusters 2 and 4. This suggests that combining multiple imaging markers allows for more precise phenotyping of COVID-19. In addition, our findings suggest that CT-based clustering could be applicable beyond COVID-19. In future pandemics, where clinical and laboratory data may be limited, CT data alone could facilitate patient phenotyping and severity stratification. Furthermore, with advancements in vaccination and treatment, the overall severity of COVID-19 has decreased². However, the utility of quantitative CT imaging in pulmonary and extrapulmonary organs has been well-established in other diseases, including COPD, interstitial pneumonia, and asthma^11,12,48. This suggests that our clustering approach may have broader applications in respiratory diseases, requiring further investigation. Finally, clustering analysis has been employed in other diseases to elucidate underlying pathophysiology. For example, in metabolic dysfunction-associated steatotic liver disease, clustering analysis has been used to elucidate different disease progression patterns and metabolic profiles, aiding in understanding pathogenesis and developing treatment strategies⁴⁹. Similarly, the clusters identified in this study may reflect distinct pathophysiological mechanisms, necessitating further research.

This study had some limitations. First, we could not evaluate the effects of vaccination or viral strains, which significantly influence COVID-19 severity and clinical presentation⁵⁰. Second, as various CT images were used from multiple models, subtle differences in requirements may have affected the analysis results. Third, we could not access the difference in treatment response among the four clusters. Fourth, there was an imbalance in the number of patients among clusters, and caution is warranted when interpreting the clinical characteristics of Cluster 4 in particular, as it included a relatively small number of patients. Future studies should address these limitations to refine further the clinical applicability of our findings.

Conclusion

This study identified four distinct clusters predicting in-hospital outcomes based on quantitative CT imaging markers. Cluster analysis incorporating multiple CT markers enabled more accurate phenotyping of COVID-19 severity. Further research is warranted to elucidate the underlying pathophysiology characterizing these clusters.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

COVID-19

Coronavirus disease 2019

CT

Chest computed tomography

COPD

Chronic obstructive pulmonary disease

VAT

Visceral fat

ESM

Erector spinae muscle

SAT

Subcutaneous adipose tissue

EAT

Epicardial adipose tissue

BMD

Bone mineral density

IPPV

Invasive positive pressure ventilation

BMI

Body mass index

Alb

Albumin

BNP

Brain natriuretic peptide

NLR

Neutrophil-to-lymphocyte ratio

Cr

Creatinine

Author contributions

Conceptualization: Shiro Otake, Shotaro Chubachi, Naoya Tanabe, and Yoshitake Yamada.Data curation: Shiro Otake, Hiromu Tanaka, Shotaro Chubachi, Ho Namkoong, Takashi Shimada, Shuhei Azekawa, Kensuke Nakagawara, Takahiro Fukushima, Mayuko Watase, and Hideki Terai.Formal analysis: Shiro Otake, Shotaro Chubachi, Tomoki Maetani, and Naoya Tanabe.Methodology: Shiro Otake, Shotaro Chubachi, Naoya Tanabe, Yusuke Shiraishi, Tomoki Maetani, Takanori Asakura, Ho Namkoong, and Yoshitake Yamada.Project administration: Shotaro Chubachi and Naoya Tanabe.Sources: Shiro Otake, Shotaro Chubachi, Hideki Terai, Mamoru Sasaki, Soichiro Ueda, Yukari Kato, Norihiro Harada, Shoji Suzuki, Shuichi Yoshida, Hiroki Tateno, and Koichi Fukunaga.Software: Naoya Tanabe, Yusuke Shiraishi, Tomoki Maetani, and Yoshitake Yamada.Supervision: Shotaro Chubachi, Naoya Tanabe, Ho Namkoong, Yoshitake Yamada, Masahiro Jinzaki, Toyohiro Hirai, Yukinori Okada, Ryuji Koike, Makoto Ishii, Akinori Kimura, Seiya Imoto, Satoru Miyano, Seishi Ogawa, Takanori Kanai, and Koichi Fukunaga.Visualization: Shiro Otake, Hiromu Tanaka, Yusuke Shiraishi, Tomoki Maetani, Shotaro Chubachi, Naoya Tanabe, and Takanori Asakura.Writing – original draft: Shiro Otake, Shotaro Chubachi, Yusuke Shiraishi, Tomoki Maetani, Naoya Tanabe, and Yoshitake Yamada.Writing – review and editing: Takanori Asakura, Ho Namkoong, Hideki Terai, Mamoru Sasaki, Soichiro Ueda, Yukari Kato, Norihiro Harada, Shoji Suzuki, Shuichi Yoshida, Hiroki Tateno, Masahiro Jinzaki, Toyohiro Hirai, Yukinori Okada, Ryuji Koike, Makoto Ishii, Akinori Kimura, Seiya Imoto, Satoru Miyano, Seishi Ogawa, Takanori Kanai, and Koichi Fukunaga.All the authors have read and approved the final manuscript.

Funding

This study was supported by the Japan Agency for Medical Research and Development (AMED) [grant numbers JP20nk0101612, JP20fk0108415, JP21jk0210034, JP21km0405211, JP21km0405217, JP21wm0325031, JP21fk0108563, JP21fk0108573, JP20fk0108452, JP21fk0108553, JP21fk0108431, JP22fk0108510, JP22fk0108513, JP22wm0325031, JP22fk0108573], Japan Science and Technology Agency (JST) CREST [grant number JPMJCR20H2], Japan Science and Technology Agency (JST) PRESTO [grant number JPMJPR21R7], and the Ministry of Health, Labour and Welfare [grant number 20CA2054].

Data availability

The datasets used and/or analysed in the current study are available from the corresponding Author upon reasonable request.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

This study was approved by the ethics committee of the Keio University School of Medicine (20200061) and related research institutions. All the participants provided informed consent. All methods were performed in accordance with the relevant guidelines and regulations.