Research design and eligibility standards
Digital medical information of sufferers with severe community-acquired pneumonia who have been admitted to the ICU of a educating hospital from January 2016 to December 2021 have been retrospectively analyzed. Inclusion standards have been: sufferers aged ≥ 18 years presenting with respiratory failure attributable to severe pneumonia necessitating ICU admission. Analysis of pneumonia relied on chest CT findings, blood C-reactive protein (CRP) ranges ≥ 4 mg/dL13, and administration of pneumonia-specific antibiotics. Severe pneumonia was outlined as assembly no less than one main or three minor standards outlined in the rules of the Infectious Illness Society of America/American Thoracic Society14. People transferred to the overall ward inside 3 days of ICU admission and people with medical situations deemed extra important than pneumonia have been excluded. Eligible sufferers have been noticed throughout hospital keep for as much as 30 days from the ICU admission date.
Medical variables
Baseline data together with age, intercourse, physique mass index, smoking historical past, and respiratory and non-respiratory comorbidities was obtained. Medical situations included standard severity-of-illness scores comparable to Acute Physiology and Persistent Well being Analysis (APACHE) II, Sequential Organ Failure Evaluation (SOFA), and SAPS II, preliminary important indicators, laboratory findings, and coverings for pneumonia. Medical outcomes included high-flow nasal cannula (HFNC), mechanical air flow (MV), extended MV, Extracorporeal Membrane Oxygenation (ECMO), extubation, tracheostomy, and in-hospital mortality.
Thoracic muscle mass measurements
Thoracic muscle mass was evaluated using chest CT scans carried out upon ICU admission of included sufferers. The thoracic muscle mass measurements embrace a complete set of 60 variables that describe numerous points of muscle amount and density inside the thoracic area. The variables are categorized into three primary teams: imply muscle space, imply muscle attenuation, and muscle quantity. Every group supplies particular insights into muscle traits.
Imply muscle space (cm2/cm) represents the common cross-sectional space of the muscle normalized by the size of the thoracic area. The imply muscle space is calculated as the overall muscle quantity divided by the size of the area of curiosity (ROI). This normalization accounts for particular person variations in physique measurement.
$$Imply ;Muscle ;Space= frac{Whole ;Muscle ;Quantity ;({cm}^{3})}{Size ;of ;ROI ;(cm)}$$
Imply muscle attenuation, measured in Hounsfield Models (HU), displays the density and high quality of muscle tissue. It’s decided by averaging the HU values of all pixels inside the segmented muscle space. The system for imply muscle attenuation includes summing the HU values of all pixels inside the muscle and dividing by the quantity of pixels.
$$Imply ;Muscle ;Attenuation= frac{sum ;HU ;values ;of ;all ;pixels}{Quantity ;of ;pixles}$$
Muscle quantity (cm3) represents the overall quantity of muscle tissue inside the specified area. It’s calculated by summing the volumes of all slices inside the ROI. For every slice, the quantity is set by multiplying the cross-sectional space by the slice thickness.
$$Muscle ;Quantity= sum (Cross-sectional ;Space left({cm}^{2}proper) occasions Slice ;Thickness left(cmright))$$
The overall muscle space in the thoracic area is the sum of all muscle areas normalized by the size of the thoracic backbone. Particular muscle tissue, comparable to the proper and left pectoralis main and minor, in addition to the proper and left erector spinae, are measured individually at totally different thoracic ranges and alongside all the thoracic backbone. For measurements particular to the erector spinae muscle group at T1–T12 ranges, attenuation values ranged from − 29 to 150 HU15. This particular vary captures the density of muscle tissue, excluding fats and different non-muscular tissues, and is beneficial in assessing the standard and composition of the muscle.
For exact analysis, we utilized 3D segmentation on chest CT scans to evaluate muscle mass of the thoracic area, specializing in particular muscle tissue such because the pectoralis main, pectoralis minor, and erector spinae. Preliminary segmentation was carried out using a radiological workstation (MEDIP, MEDICALIP, Seoul, South Korea. https://medicalip.com/medip/), which mixed 3D quantity rendering and multiplanar picture reformatting. This workstation employs a semi-automatic segmentation methodology using a graph lower algorithm16. It’s identified for having precision in defining muscle boundaries with the steering of an skilled radiologist. To additional improve the accuracy and effectivity of muscle segmentation, we integrated a custom-developed deep studying AI mannequin tailor-made for 3D convolutional neural network-based segmentation. This mannequin is particularly educated to establish and delineate focused muscle teams. Detailed details about the muscle segmentation community, together with the coaching knowledge, coaching pipeline, and check efficiency, has been summarized in Supplementary Data S1.
Furthermore, AI software program (DeepCatch, MEDICALIP, Seoul, South Korea. https://medicalip.com/DeepCatch/) was used for measuring whole muscle mass at numerous segments of the backbone, particularly at T4, T12, and from T10 to T1217. A complete measurement from T1 by T12 was additionally carried out. This AI software program excels in offering detailed visualizations and quantifications of muscle mass by superior imaging analysis.
Clustering stability
To judge the reproducibility and stability of our clustering analysis approach, we carried out a sensitivity analysis using random subsampling. We generated 100 random subsets of the information, every containing 80% of the unique pattern measurement. For every subset, we standardized the information and carried out k-means clustering with ok = 3, sustaining the identical parameters as the unique analysis. We calculated the Adjusted Rand Index (ARI) for every pair of clustering outcomes throughout the subsets to evaluate the similarity of the clustering buildings.
Outcomes
The first consequence was extubation. The secondary consequence was in-hospital mortality over a 30-day remark interval.
Predictive mannequin evaluation
We employed a logistic regression mannequin to foretell profitable extubation and in-hospital mortality primarily based on thoracic muscle mass measurements. The whole dataset was analyzed with out splitting it into coaching and testing units. This strategy was chosen to carry out an exploratory analysis for preliminary mannequin evaluation and to maximise knowledge utilization, significantly in evaluating the feasibility of using thoracic muscle mass knowledge for predictive modeling.
Statistical analysis
Okay-means clustering was carried out with 60 variables representing profiles of thoracic muscle mass. Clusters have been visually depicted by uniform manifold approximation and projection (UMAP) and principal part analysis (PCA) plots. Categorical variables have been analyzed using the chi-squared check, whereas steady variables have been analyzed using the unbiased t-test or Mann–Whitney U check. A chi-square check for linear development was carried out to guage scientific outcomes in response to clusters. Kaplan–Meier curves have been employed to investigate time to extubation and in-hospital mortality occasions. Hazard ratios for extubation and in-hospital mortality have been estimated using Cox proportional hazards fashions. We assessed the general efficiency using the world underneath the receiver working attribute curve (AUC) and in contrast fashions using the Delong methodology18. Moreover, accuracy, specificity, sensitivity, precision, and recall have been evaluated. Statistical significance was set at p < 0.05, alongside 95% confidence intervals. Statistical analyses have been carried out using R statistical software program model 4.1.2 (R Basis, Vienna, Austria).
Ethics
This examine adopted moral tips outlined in the Declaration of Helsinki of 1975. The Institutional Evaluate Board (IRB) of Boramae Medical Middle accepted the examine protocol and waived the requirement for knowledgeable consent from examine individuals to entry their digital medical information (IRB No. 10-2021-110).