Metabolomic Biomarkers Predict Long-term Physical Function in Survivors of Acute Respiratory Failure

Metabolomic Biomarkers Predict Long-term Physical Function in Survivors of Acute Respiratory Failure | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Metabolomic Biomarkers Predict Long-term Physical Function in Survivors of Acute Respiratory Failure Adeyeye I. Haastrup, Justin T. Roberts, Sheetal Gandotra, Emily M. Hartsell, and 9 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7394034/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 21 Nov, 2025 Read the published version in Respiratory Research → Version 1 posted 8 You are reading this latest preprint version Abstract Introduction: Acute respiratory failure (ARF) often leads to post-intensive care syndrome, including persistent physical impairments after ICU discharge. Emerging evidence suggests that mitochondrial bioenergetic dysfunction, detectable through metabolomic profiling, may contribute to poor recovery. Methods: We performed a retrospective study comprising of untargeted metabolomic profiling using ultrahigh performance liquid chromatography–mass spectrometry (UHPLC-MS) on serial serum samples from 70 ARF patients taken at ICU admission, during hospitalization and at discharge. Physical function was assessed post-discharge using the Short Physical Performance Battery (SPPB). Correlation and logistic regression analyses were performed to identify metabolomic predictors of six-month physical function outcomes. Results: Patients with poor SPPB scores exhibited dysregulation in bioenergetic metabolite levels, as well as fatty acid oxidation, glycerophospholipid metabolism, bile acid biosynthesis and amino acid metabolism. These metabolic changes were not explained by initial disease severity (APACHE III scores) or comorbidities. In contrast, several metabolites measured at discharge were predictive of SPPB scores with an AUROC of 0.88 after cross validation. Conclusion: Our findings highlight persistent metabolic dysfunction at discharge, particularly in pathways related to bioenergetics. To our knowledge, this is the first study to employ a metabolite-based machine learning model to predict ARF survivors physical function outcomes using serum metabolites measured at discharge. Further insights on dysregulated pathways suggest that nutritional interventions targeting these metabolic pathways, such as supplementation with β-alanine, could potentially improve post-ICU recovery outcomes. Acute respiratory failure post intensive care syndrome physical function metabolite feature selection for patient classification logistic regression. Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Acute respiratory failure (ARF) is a life-threatening condition characterized by inadequate blood oxygenation with or without hypercapnia [1,2]. It is often associated with critical conditions such as sepsis, chronic obstructive pulmonary disease (COPD), and acute respiratory distress syndrome (ARDS), and imposes a substantial socioeconomic burden on affected patients and society as a whole [3,4]. A multinational study revealed that more than 50% of critically ill patients in the ICU have or will develop ARF, which places them at a 40% risk of mortality [5]. Survivors of ARF often face multiple physical and cognitive difficulties that are collectively termed post-intensive care syndrome (PICS) after ICU discharge [6–9]. About 46% of ARF survivors are at high risk of readmission or dying within 3–4 months following initial ICU discharge which may be due to PICS-related disabilities [10,11]. Persistent physical disabilities after discharge from the ICU [6,12,13], reflecting impairments in muscle strength, mobility, and overall function long after the acute illness [14], are core components of PICs. In survivors of COVID-19-related ARDS, the incidence of physical function impairment is approximately 74% at 1 year after ICU treatment [15]. The quality of life decrement for these patients is influenced significantly by various factors, including underlying health conditions, demographics, ICU treatment, and the medical, pharmacological, and rehabilitative follow-up treatments [16–19]. The Short Physical Performance Battery (SPPB) is a tool commonly used to measure physical performance in the elderly [20]. It has been shown to correlate with various important health outcomes, including the ability to perform daily activities, risk of hospital admission or readmission, overall functional status, and mortality rates in critically ill and elderly subjects [21,22]. While the etiology of ARF is reasonably well understood, pathophysiologic mechanisms contributing to persistent poor physical performance in post-ICU patients, as well as reliable tools to predict long-term functional outcomes remain elusive [12,23]. The long-term quality of life outcomes for ARF survivors, identified as a key research priority by critical care professional societies and pulmonary physicians, emphasizes the urgent need for research focused on delineating the pathophysiologic changes that occur at the molecular level [24,25]. Metabolomics is increasingly utilized to profile interactions at a system level in patients with critical illness [26,27]. This approach may enable a comprehensive and holistic understanding of the disease process as well as improvements in diagnosis, assessing response to therapy, monitoring disease progression, pathological mechanisms, biomarker, and pharmacologic target discovery [28]. Our previous studies identified bioenergetic dysregulation as a key factor in non-survivors of sepsis, ARF, and acute kidney injury (AKI) [29–34]. The prominently affected pathways in non-survivors were bioenergetic, specifically NAD⁺ metabolism and β-oxidation, which are essential for mitochondrial bioenergetics, and these were accompanied by disruptions in acetylcarnitine, glutathione, bile acid, steroid, and fatty acid metabolism[30]. Building on this, because physical exercise performance is inherently linked to mitochondrial bioenergetics [35,36], we postulated that the dysregulation in bioenergetic pathways may follow a continuum, with varying degrees of severity correlating with a range of outcomes. For example, those with the most pronounced metabolic disruptions did not survive, while survivors with poor outcomes experience milder but still impactful bioenergetic dysregulation, that potentially impairs physical performance. In this study, we hypothesized that poor physical function in ARF survivors is associated with, and predicted by, bioenergetic dysfunction, as reflected by metabolomic and mitochondrial biomarkers. Methods This retrospective study involved ARF patients enrolled in the Standardized Rehabilitation for ICU Patients with Acute Respiratory Failure clinical trial at Wake Forest Baptist Medical Center in North Carolina (ClinicalTrials.gov Identifier, NCT00976833; registration date, 2009-09-11). Institutional review board approval was obtained, and informed consent was provided by patients or their legal representatives. ICU patients were enrolled based on a clinical diagnosis of ARF made by the attending physicians, following the inclusion and exclusion criteria previously outlined [30,37]. All patients were treated according to standard of care practices, comprising a variety of pharmacological therapies. To ensure comparability between the groups, the patients were matched for age, race, and sex. Demographics of the selected patients are summarized in Table 1 . Global metabolomics was performed on patients' serum samples at three time points: ICU admission (n = 70), five days post-admission (n = 40), and ICU discharge (n = 69). Before blood collection from a central line, the catheter was flushed with sterile saline and a waste sample was drawn to remove residual fluids or anticoagulants, ensuring uncontaminated samples. Samples were collected at enrollment, day 5 post-enrollment, and at ICU discharge. The smaller number of samples at day 5 and at discharge reflects both study design and patient availability. We anticipated that this timepoint would provide limited clinical utility beyond showing temporal trends, so we prioritized enrollment and discharge samples to maximize predictive power. Furthermore, several patients in both good and poor outcomes were discharged prior to day 5. There was one sample at discharge that was not measured by Metabolon and therefore removed from the analysis. At discharge, and at two-, four-, and six-months following ICU admission, we assessed physical function using the SPPB, an objective measure that evaluates gait speed, balance, and lower extremity strength [38]. Serum Metabolomics : Extraction and quantification of serum metabolites were performed by Metabolon Inc [30]. Briefly, 100ul of each sample was prepared with recovery standards using the automated MicroLab STAR® system (Hamilton Company). Sample runs were performed on C18 and HILIC columns coupled to a Thermo Scientific Q-Exactive high-resolution/accurate mass spectrometer with the orbitrap mass analyzer operated at 35,000 mass resolution in negative and positive ionization modes. Metabolon's hardware and proprietary software were used for spectral data extraction and compound identification. Artifacts and background noise were removed. Peaks were quantified using the area-under-the-curve method. Specifically, identifications are based on a narrow retention index, mass accuracy within ± 10 ppm, and MS/MS spectral matching to authentic standards (forward and reverse scores). The scaled intensity of metabolites was used to provide semi-quantitative metabolite values, wherein each value quantifies the relative concentration of a metabolite in a sample. The mass area was normalized to account for sample volume, blanks, quality controls, and internal standards. Variance filtering for repeatability (using QC samples) was performed using internal standards and endogenous metabolites, with relative standard deviation thresholds of 3% and 9%, respectively. Metabolomic Data Processing and Statistical Comparisons : Metabolites with greater than 30% missing points were excluded from further processing [39,40]. Missing data points were imputed with mechanism-aware imputation algorithm [40]. A Shapiro-Wilk test performed on the data before and after imputation gave p-value < 0.05, indicating non-normal distribution of the metabolite data. To improve signal-to-noise ratio and enhance downstream modeling, metabolites with a relative median absolute deviation less than 0.25, representing near-constant features with limited statistical power, were excluded [41,42]. Metabolite intensities were transformed to log scale in order reduce skewness. To explore differences in metabolite levels between functional outcome groups, we applied the Wilcoxon rank sum test (two-sided) to compare survivors with poor (SPPB ≤ 6) and good (SPPB ≥ 7) scores. Metabolites with p < 0.05 were considered to show the strongest differences and are hereafter referred to as “differential metabolites”. We explored false discovery rate (FDR) correction using multiple methods including the Benjamini-Yekutieli procedure; however, due to the highly discrete distribution of p-values, modest sample size and limited variability among test statistics, the adjustment yielded uniformly non-significant results. As such, FDR-adjusted results were not used for inference. Metabolomic profiles were analyzed separately at three timepoints: day 1, day 5 post admission, and at hospital discharge. Multivariate hierarchical clustering of differential metabolites was performed to assess temporal changes in metabolomic profiles between good and poor physical function groups across the three time points. The data were analyzed using custom-developed scripts in RStudio version 4.4.1 [43]. Statistical analysis of the demographic data was conducted using Wilcoxon rank-sum test for continuous variables and Fisher’s exact test for categorical variables. A p-value < 0.05 was considered significant when comparing poor and good physical performance groups. Chemical Similarity Enrichment, Pathway and Meta-Analysis : Chemical similarity enrichment analysis was performed using ChemRICH, which enabled us to identify classes of compounds that were up- or downregulated. ChemRICH also offers the advantage of identifying clusters of chemically similar metabolites, revealing potential new metabolite sets that may be relevant [44]. Pathway analysis was performed using MetaboAnalyst 6.0 focusing on differential metabolites to identify specific dysregulated pathways in subjects with poor physical function [45,46]. A meta-analysis was conducted to evaluate the association of self-declared race (white or black) and gender (male or female) with differential metabolites that were either upregulated or downregulated. In addition, Fisher's Exact Test was used to assess the association between physical function outcomes, categorized as good vs. poor SPPB, and clinical parameters including APACHE III score, body mass index (BMI), mean arterial pressure (MAP), heart rate (pulse), respiratory rate, fraction of inspired oxygen (FiO₂), and arterial partial pressure of carbon dioxide (PaCO₂), as well as comorbidities present at admission. Metabolite Feature Selection, Classification, and Model Validation : As indicated subsequently, the highest number of differential metabolites and dysregulated pathways was observed in the poor SPPB group at discharge. To streamline relevant features for sample classification, we applied a filtering process on differential metabolites measured at discharge. Log transformed data were mean-centered and scaled to unit variance. Using the publicly available online version of MetaboAnalyst™ software, a Partial Least Squares Discriminant Analysis (PLS-DA) was performed to classify the patients into the respective SPPB categories, the classification being executed utilizing the biomarker candidates comprising only the differential metabolites. Additionally, a supervised machine learning, Random Forest classification model was implemented to rank biomarker candidates features based on their discriminatory power, using the randomForest and caret R packages [47,48]. Ten metabolites identified at this stage were subjected to Spearman correlation analysis, to eliminate redundant features exhibiting high collinearity or duplicative discriminatory effects. A correlation threshold of 0.8 was defined, and biomarker pairs exhibiting absolute correlation values exceeding this threshold were identified as highly correlated features. Furthermore, a Bayesian logistic regression model incorporating leave-one-out cross-validation was utilized (rstanarm and loo packages in R) to assess the impact of inclusion or exclusion of metabolites on model performance [49,50]. Based on this iterative evaluation, seven metabolite biomarkers identified as the most predictive of patients' physical function were selected for standard logistic regression analysis with 5-fold cross-validation. Given the strong performance of the standard model, we further evaluated the predictive power of these metabolites using a Bayesian logistic regression framework for comparison. To evaluate the generalizability of the model to unseen data, a 5-fold cross-validation procedure was conducted. To validate the robustness and stability of the model across variable data partitions, a repeated k-fold cross-validation procedure was implemented, through twenty repetitions of 5-fold cross-validation. The execution parameters included five independent Markov chains, each with 2,000 iterations and a fixed random seed of 12345 for reproducibility. Results To determine if bioenergetic metabolites are predictive of six-month physical function in survivors of ARF, we performed broad spectrum metabolomic analysis of serum samples collected at enrollment (day 1), day 5 post enrollment, and patient discharge. Seventy patients were selected from the Standardized Rehabilitation for ICU Patients with Acute Respiratory Failure cohort. Patients were matched for age, race, and sex (Table 1 ). Half the patients were determined to have a “good” physical performance, (SPPB ≥ 7), while the other half had a “poor” physical performance (SPPB ≤ 6). Demographics of the selected patients are summarized in Table 1 . Table 1 Patients Demographics Variable Good Poor P-value n-value 35 35 SPPB score 10 [8.5–10] 3 [0–5] < 0.001 a APACHE III score 70 [57.5–82.5] 69 [60.5–79] 0.764 a MAP (mmhg) 65 [56–93] 68 [57–100] 0.477 a Pulse (bpm) 103 [85–123] 106 [85.5–114] 0.577 a Respiratory rate (breaths/min) 24 [20.5–30.5] 22 [19.5–28.5] 0.303 a PCO 2 (mmhg) 44 [34-52.25] 37.6 [32.5–43.7] 0.127 a PAO 2 (mmhg) 95 [77–121] 91.6 [75.05–112.5] 0.703 a FIO 2 (mmhg) 0.5 [0.4–0.6] 0.6 [0.4–0.7] 0.386 a BUN (mg/dl) 19 [15–29] 18 [11.5–28] 0.577 a Serum albumin (g/dl) 3 [2.6–3.3] 2.7 [2.3–3.2] 0.127 a Serum bilirubin (mg/dl) 0.8 [0.6–1.05] 0.9 [0.55–1.15] 0.991 a Blood glucose (mg/dl) 137 [100–204] 132 [116.5-164.5] 0.707 a BMI (kg/m 2 ) 30.27 [26.6-36.33] 29.67 [23.07–34.35] 0.280 a Age (years) 62 [54-69.5] 59 [54-71.5] 0.986 a LOS (days) 9 [7-11.5] 12 [8.5–17] 0.019 a Sex 1 b Female 18 (51.4%) 18 (51.4%) Male 17 (48.6%) 17 (48.6%) Race 1 b Black 11 (31.4%) 11 (31.4%) White 24 (68.6%) 24 (68.6%) Nutrition 1 b No 30 (85.7%) 29 (82.9%) Yes 5 (14.3%) 6 (17.1%) Home oxygen 1 b No 25 (71.4%) 25 (71.4%) Yes 10 (28.6%) 10 (28.6%) Dialysis 1 b No 31 (88.6%) 33 (94.3%) Yes 4 (11.4%) 2 (5.7%) Physical therapy 1 b No 15 (42.9%) 15 (42.9% Yes 20 (57.1%) 20 (57.1%) Mass spectrometry analysis was performed by Metabolon Inc. as described in the methods. In total, 758 named biochemicals and 151 unnamed biochemicals were detected (total: 909 biochemicals). After removing near-constant features and restricting to named compounds, 742 biochemicals were retained for analysis. Samples identified as outliers by Metabolon Inc. were removed from the statistical analysis, reducing the number of samples to 66, 39, and 64 on day 1, day 5, and at discharge, respectively. Further analysis was performed on named biochemicals. Univariate and Multivariate Analysis Results : On day one, 21 serum metabolites concentrations were identified as having the strongest difference (p < 0.05) between outcome groups. Patients with poorer physical outcomes exhibited a marked reduction in bioenergetic-related metabolites, including trigonelline, a metabolite of NAD⁺, which plays a key role in cellular energy metabolism (Fig. 1 A) [51]. At day 5 post-admission, 21 serum metabolites differentiated patients with poor and good physical performance. Androgens play crucial roles in bioenergetics [52,53]. Amongst the differential metabolites, five androgen-related metabolites were lower in patients who had poor physical function (dehydroepiandrosterone sulfate [DHEA-S], androsterone sulfate, epiandrosterone sulfate, 5-alpha-androstan-3beta,17beta-diol disulfate, and androstenediol [3alpha, 17alpha] monosulfate) (Fig. 1 B). Remarkably, 67 differential metabolites distinguished patients with poor versus good physical function at discharge. Among ARF survivors with good SPPB scores, 22 differential metabolites were decreased at discharge, compared to only 6 on day 5 and 4 on day 1. In contrast, survivors with poor SPPB scores showed 45 differential metabolite with decreased concentrations at discharge, 15 at day 5, and 17 at day 1. This highlights a noticeable shift in the metabolomic profile by the time of hospital discharge. Bioenergetic metabolites such as hydroxyl fatty acids, fructose, glycerides, alpha-ketoglutarate, and glycerol-3-phosphate were reduced in patients with poor physical function (Fig. 1 C). Although the primary aim of this study was to identify predictive metabolite biomarkers and develop an FDR-independent predictive model, initial analyses revealed several metabolites with nominal significance (p 0.05). Multivariate analysis of the metabolite trends between the good and poor SPPB score groups across time point showed some temporal differences reflecting dynamic metabolic shifts across the three time points (Additional file1: figure S1 ). Chemical Similarity Enrichment and Pathway Analysis Results : We performed chemical and pathway enrichment analyses to evaluate chemical similarities among enriched metabolites, identify trends in metabolic pathway alterations during hospitalization, and highlight potential pathways that may be targeted for therapeutic intervention. Analysis of the metabolomic panel on day 1 reveals disruption in the catabolism of valine, leucine, and isoleucine in ARF survivors with poor SPPB scores (Fig. 2 A). These branched-chain amino acids are essential for muscle metabolism and mitochondrial energy production [54]. Pathway analysis further underscores slight perturbations in cysteine and methionine metabolism. ChemRICH analysis highlights the decreased keto acids (branched-chain amino acid metabolites) and increased cholic acids (bile acids) which corresponds with metabolic pathways identified as impacted by the pathway analysis with MetaboAnalyst (Additional file 1: Figure S2 A). On day five, pathway analysis demonstrates dysregulation in both primary bile acid biosynthesis and amino acid metabolism in group of ARF survivors with poor SPPB scores (Fig. 2 B). Similar to day one, ChemRICH analysis only revealed decreased levels of keto acids associated with group with poor SPPB scores (Additional file 1: Figure S2 B) At discharge, pathway analysis documented alterations in the tricarboxylic acid (TCA) cycle, suggesting disrupted cellular energy production with potential adverse effects on physical recovery in patients with poor SPPB scores. Alanine, aspartate and glutamate metabolism and glycerophospholipid metabolism are also key pathways dysregulated at discharge (Fig. 2 C). Based on the ChemRICH analysis, carnitines emerged as the most elevated, while saturated fatty acids were the most decreased class of metabolites in patients with poor physical performance at the discharge time point (Additional file 1: Figure S2 C). Although several pathways, including "Valine, leucine, and isoleucine degradation" and "Primary bile acid biosynthesis," showed as dysregulated at earlier time points (ICU admission and day 5), their biological impact scores were limited, indicating subtle metabolic shifts. In contrast, pathway impact scores notably increased at discharge (Fig. 2 ). Meta-Analysis : We were interested in determining if patient demographics (age, race, and sex) were associated with metabolomic changes. Overall, there was no association with age. On day 1, the only differential metabolite that was slightly downregulated in Black patients was 3-indoxyl sulfate. Females exhibited higher levels of the bile acid taurocholate at both day 1 and day 5 compared to males. 1-palmitoyl-2-palmitoleoyl-GPC, was also differentially elevated in White patients than in Black patients on day 5. At discharge, the metabolite profile showed no gender-based differential abundance. However, four metabolites were elevated in Black patients and three in White patients. Black patients showed higher levels of bile acids (taurochenodeoxycholate, hyocholate), threonate, and the peptide HWESASXX, which has been linked to worse outcomes in COPD [55]. In contrast to Black patients, White patients showed elevated levels of carnitine conjugates of medium and long chain fatty acids [(myristoleoylcarnitine (C14:1), palmitoleoylcarnitine (C16:1), 5-dodecenoylcarnitine (C12:1))], metabolites related to β-oxidation and mitochondrial function. We were also interested in the potential influence of co-existing illness on the metabolic profile of the ARF patients that may be related to poor or good physical function. No significant associations were observed between patient physical function and comorbidities, APACHE III scores, or other clinical measures such as mean arterial pressure (Additional file 1: Figure S3 ). Patient Classification Modeling : Given that the most differential serum metabolomic differences were observed at patient discharge compared to day 1 and day 5, we focused on this time point using the data of 30 participants with poor SPPB scores and 34 with good SPPB scores (Additional file 1: Figure S4 A – S4C). We developed classification models to assess whether metabolomic profiles could anticipate the six-month SPPB scores. To identify the most relevant metabolites, we applied PLS-DA, Spearman correlation matrix, random forest feature selection, and evaluated using Bayesian leave-one-out cross-validation (Additional file 1: Figure S4 ). Seven predictive metabolite biomarkers were selected for further analysis: cystine, glycerol 3-phosphate, hyocholate, 2-hydroxy-3-methylvalerate, taurocholenate sulfate, 1-oleoyl-2-docosahexaenoyl-GPC (18:1/22:6), and ximenoylcarnitine. These were subsequently used in a standard logistic regression model to assess their predictive performance. The standard logistic regression model achieved an accuracy of 0.86, along with a strong area under the curve (AUC) of 0.94. Following 5-fold cross-validation, the model continued to perform well, with an accuracy of 0.75 and an AUC of 0.86 (Additional file 1: Figure S5 ). These results suggest good discriminative ability without evidence of overfitting and support its potential to generalize to independent validation cohorts. To compare with standard logistic regression, we assessed the predictive capacity of the seven metabolites using Bayesian logistic regression framework. The intercept’s posterior distribution of the posterior plot, centered near zero, suggests baseline equal odds of 'Poor' versus 'Good' physical condition when metabolites are average (Fig. 3 A). A slightly positive posterior indicates a mild bias toward predicting 'Good' SPPB scores. The posterior distributions for cystine, glycerol 3-phosphate, and 2-hydroxy-3-methylvalerate lie predominantly above zero, indicating that higher levels of these metabolites credibly increase the odds of a ‘Good’ SPPB score. In contrast, the posteriors for hyocholate, taurocholenate sulfate, 1-oleoyl-2-docosahexaenoyl-GPC, and ximenoylcarnitine lie below zero, suggesting that elevated concentrations are associated with poorer physical function. All metabolites present moderate posterior distributions with 95% credible intervals clear of zero, underscoring consistent associations in predicting physical function (Fig. 3 A). The posterior predictive plot (Additional file 1: Figure S6 A) shows a close correspondence between simulated and observed outcomes, with the predictive intervals tightly enveloping the empirical data. This concordance indicates that the Bayesian model faithfully captures the underlying data structure and produces reliable predictions. The confusion matrix and ROC curve of Bayesian model (Fig. 3 B and 3 C) indicate strong model-fit for classifying patients into 'Good' and 'Poor' SPPB groups [56]. The trace plot indicates convergence to the target distribution, suggesting well-sampled posterior distributions (Additional file 1: Figure S6 B). Low autocorrelation in MCMC chains implies efficient Bayesian sampling, with reliable posterior estimates (Additional file 1: Figure S6 C) [57]. Patient classification under the Bayesian framework was further assessed using 5-fold and repeated k-fold cross-validation (100 model evaluations), achieving a robust AUC of 0.88 (Fig. 4 ). Collectively, these results underscore strong predictive associations between serum metabolites measured at hospital discharge and subsequent patient physical function. The consistently high AUC values and accuracy across multiple validation approaches indicate that the selected metabolite panel provides stable and effective predictive utility for future patient physical performance outcomes. The line plot illustrates the trends of the seven predictive metabolites biomarkers and β-alanine at key time points during hospitalization, day 1 of ICU admission, day 5, and at discharge (Fig. 5 ). Discussion Based on our prior work in ARF patients, we hypothesized that the most pronounced metabolomic differences between patients with good and poor physical function would occur at ICU admission, reflecting the acute severity of illness. However, our findings revealed a surprising pattern: despite clinical improvement by the time of discharge, the number of differential metabolites was threefold higher at discharge compared to ICU admission or mid-hospitalization (Additional file 1: Figure S7). The only metabolite that consistently differed at all three time points between patients with poor and good physical function was the bile acid taurochenodeoxycholate, showing higher mean and median values in those with poor SPPB test results. Additionally, bile acids were generally elevated across all time points in patients with poor physical performance, suggesting a consistent dysregulation of bile acid metabolism in this group. These findings are further supported by hierarchical clustering of multivariate metabolite profiles, which revealed some noticeable patterns between good and poor SPPB groups across time points, especially in differential metabolites related to bile acid and phosphatidylethanolamines metabolism and bioenergetics (Additional file 1: Figure S1 ). ChemRICH and pathway analysis (using MetaboAnalyst) identified similar, though slightly varied dysregulated metabolomic alterations at each time point. These findings suggest that while the overall metabolic disruptions are consistent, the specific pathways and metabolites involved may shift as the patient progresses through the phases of hospital care and recovery. Our previous research demonstrated a correlation between acute mortality in ARF patients and elevated carnitine levels alongside mitochondrial-related bioenergetic dysfunction [30]. Notably, ChemRICH analysis of metabolites at discharge revealed that carnitines constituted the most enriched biochemical set in patients with poor SPPB scores, supporting the association between carnitine upregulation and worse outcomes in critically ill patients (Additional file 1: Figure S2 C) [30]. Patients with poor physical function showed reduced lipid and fatty acid levels, and elevated bile acids (Fig. 1 C; Additional file 1: Figure S2 A, S2 C), consistent with prior studies linking low lipid levels in critical illness to worse outcomes [58,59]. These metabolomic differences suggest a shift in bile acid homeostasis, potentially reflecting altered hepatic metabolism, although serum albumin and bilirubin levels did not differ significantly between groups (Table 1 ) [60–63]. At the discharge time point, disruption in glycerophospholipid metabolism in the poor SPPB group may impair membrane lipid turnover, which is critical for maintaining cellular integrity and membrane fluidity. Similar reductions in glycerophospholipid metabolism have been previously observed to correlate with poor outcomes in patients with COPD (Fig. 2 C and Additional file 1: Figure S1 ) [64]. Nutritional status variations also could affect the metabolic disparities between patients with high and low SPPB scores [65]. The lower fatty acid and lipids levels observed with poor SPPB scores suggest that the increased bile acid production may be a compensatory metabolic response, potentially important for enhancing lipid digestion and absorption. On day 1, 3-indoxyl sulfate, a tryptophan-derived metabolite was downregulated in Black ARF survivors and has been implicated in microbial–host interactions, gut homeostasis, and uremic toxicity [66,67]. Moreover, a diet modifiable metabolites, 1-palmitoyl-2-palmitoleoyl-GPC, was differentially elevated in White patients than in Black patients on day 5 [68]. Given that only few ARF patients in the cohort received nutritional therapy, and considering the known impact of nutritional status on both short- and long-term outcomes in critical illness, metabolomic profiling may offer valuable insights into abnormal metabolic signatures. These insights could inform personalized nutritional strategies to improve long-term functional outcomes in ARF patients [69,70]. It is also noteworthy that the APACHE III score showed no significant association with poor physical performance. While this clinical metric provides an overall assessment of patient condition, it may not capture the nuanced biological variability that influences physical function in critically ill patients. Likewise, other important variables such as physical therapy, home oxygen use, and dialysis have no significant association with poor or good physical function. The lack of significant association in aforementioned parameters with the physical function status may be due to modest sample size used in this preliminary study. On the contrary, patients with longer ICU stays were significantly more likely to have poor physical function outcomes (Table 1 ). We observed decreased levels of β-alanine at discharge in patients with poor SPPB scores (Fig. 1 C and 2 C). β-alanine is a central metabolite serving as a precursor in various key metabolic processes, including the synthesis of carnosine, a dipeptide of β -alanine and histidine, which supports muscle metabolism [71–75]. β-alanine intake could thus potentially benefit ARF patients prone to poor physical performance and thus seem to warrant additional research into its clinical applications [71,72,74–76]. The strong performance of both logistic regression and Bayesian logistic regression models demonstrates that metabolomic profiling at the time of ICU discharge can effectively predict patients’ subsequent physical performance (Additional file 2: Table S1 ). While frequentist logistic regression shares a similar foundational framework with its Bayesian counterpart, the Bayesian approach demonstrated superior performance during cross-validation, achieving an AUC of 0.88 compared to 0.86 with standard logistic regression. Moreover, the Bayesian model yielded fewer false positives and false negatives in sample classification at cross validation (Fig. 4 and Additional file 1: Figure S5 ). To our knowledge, this represents the first statistically grounded machine learning model that leverages metabolomic data to classify and predict future physical function outcomes in survivors of ARF. Some metabolites with predictive utility (2-hydroxy-3-methylvalerate, hyocholate, cystine, and β-alanine) showed temporal trends and partial separation between good and poor SPPB groups at day 1 or day 5, but none were considered differential (nominal p > 0.05) at these time points (Fig. 5 and Additional file 1: Figure S7). Independent validation of the predictive performance of the Bayesian logistic regression model is necessary to further strengthen the case for the application of these metabolites to predict PICS of discharged ARF patients. The biomarkers identified as predictive of physical function in this study are FDR-independent, as they were selected based on multivariate modeling rather than univariate significance [77,78]. A limitation of this study is that the metabolites identified differential (p 0.05) (see Additional File 3). This may be partly attributable to the modest sample size. As part of our next step, determining absolute metabolite concentrations through calibration to internal and external standards in a targeted metabolomics approach will help minimize variability in metabolite quantification and with increased sample size and statistical power, thereby strengthening the validation of these findings [79]. This pilot study was designed to maximize the use of available resources and to guide the identification of metabolites for focus in future, more targeted investigations. We were also limited by the number of patients that had the final 6-month SPPB scores. Future prospective studies may need to utilize different tools for determining physical function, such as activity tracker watches that do not require direct patient interaction after patient discharge. While the predictive modeling and cross-validation results are promising, the absence of an independent validation cohort limits our ability to assess the model’s generalizability and stability. Increased hospital length of stay among ARF survivors enrolled in this study was significantly associated with poor SPPB scores, indicating worse physical function. This variability in discharge timing may have introduced some heterogeneity into the metabolomic findings, particularly if longer hospitalization reflects differing stages of recovery or persistent metabolic impairment. Importantly, the relatively longer LOS in survivors with poor physical performance aligns with our postulate of a continuum of bioenergetic dysregulation. Non-survivors in our prior studies exhibited severe mitochondrial dysfunction and global metabolic collapse, the survivors with diminished physical function in this cohort may represent an intermediate phenotype, marked by sub-lethal, bioenergetic deficits that impair recovery [29–34]. All patients in the study cohort received standard of care which included various medications, that may have some influence on the metabolomic profile. We did not stratify outcomes based on the underlying etiology of ARF (infectious vs. non-infectious), which may have contributed to variability in the observed metabolomic and functional outcomes. This study was retrospective; future prospective studies are necessary to confirm the clinical utility of the predictive model. Targeted metabolomics, particularly for NAD⁺, glycolysis, and mitochondrial redox pathways, may improve sensitivity and specificity in future analyses. Longitudinal monitoring of post-discharge metabolomic profiles would help determine whether these pathways remain significantly altered in patients with poor outcomes and could also assess the impact of interventions such as diet and physical therapy on metabolic recovery. In the next phase of our research, we will conduct prospective validation in independent ARF cohorts across multiple centers with increased sample size. These studies will also incorporate socioeconomic factors and data on access to, and engagement in, physical therapy, thereby providing a more comprehensive understanding of the determinants of physical recovery after ICU discharge. Conclusions Metabolomic profiling of bioenergetic-related metabolites at ICU discharge can distinguish ARF survivors with good versus poor physical performance. We propose that serum metabolomic analysis at discharge offers actionable clinical insight into a patient’s metabolic status and recovery trajectory. Quantitative targeted measurement of key bioenergetic and bile acid metabolites may enable risk stratification, inform nutritional or rehabilitative interventions, and personalize post-ICU care. While future prospective studies are required to validate the accuracy and clinical utility of this model, the predictive metabolomic panel identified in this study holds strong potential as a prognostic tool to identify patients at high risk for poor physical recovery and to guide targeted interventions aimed at improving long-term outcomes. Abbreviations SPPB short physical performance battery ARF acute respiratory failure UHPLC-MS ultra-high performance liquid chromatography–mass spectrometry ICU intensive care unit COPD chronic obstructive pulmonary disease ARDS acute respiratory distress syndrome PICS post-intensive care syndrome PLS-DA partial least squares discriminant analysis APACHE Acute Physiology and Chronic Health Evaluation TCA tricarboxylic acid cycle ROC receiver operating characteristic curve AUC area under the curve. Declarations Availability of Data and Materials The dataset supporting the conclusions of this article is included within the article and its additional files. The code used to process and analyze the metabolomics data is publicly available at https://github.com/RNABioUSA/arfqol-metabolomics. Ethics approval and consent to participate: This retrospective study involved ARF patients enrolled in the Standardized Rehabilitation for ICU Patients with Acute Respiratory Failure clinical trial at Wake Forest Baptist Medical Center in North Carolina (ClinicalTrials.gov Identifier, NCT00976833; registration date, 2009-09-11). The Wake Forest IRB approval is in accordance with the Declaration of Helsinki. Institutional review board approval was obtained, and informed consent was provided by patients or their legal representatives. Consent for publication Not applicable. Competing Interest: We developed a predictive model using a Bayesian framework that incorporates seven key metabolite biomarkers to estimate future physical function in critically ill patients following hospital discharge. To support future clinical translation, a provisional patent application has been filed to secure intellectual property rights for this biomarker-based prediction method. Acknowledgements Not applicable. Authors' contributions AIH, JTR, GTD, RGB data analysis, manuscript preparation. LDP Clinical data collection and data analysis. SG, DCF (D. Clark Files) clinical data analysis and manuscript preparation. EMH, VMP, TS manuscript preparation. DCF, PEM, MNG, and RJL conceptualized the idea, generated the data, guided the development, funding, and revision of the manuscript. Funding NIH Grants: 1UL1TR001417; 1KL2TR003097; 1 R21 NR019338-01. Center for Lung Biology Frederick P. Whiddon College of Medicine: Gary and Susan Godwin Emerging Scholars Endowed Award; Murray Bander Faculty Development Award. Frederick P. Whiddon College of Medicine Dean's Predoctoral Fellowship Award. This work was supported in part by an Early Career Investigator Award from the American Thoracic Society (JTR). References Hanley ME, Bone RC. Acute respiratory failure. Pathophysiology, causes, and clinical manifestations. Postgrad Med. 1986;79(1):166–76. Summers C, Todd RS, Vercruysse GA, Moore FA. Acute Respiratory Failure. In: Perioperative Medicine. Amstardam: Elsevier; 2022. p. 576–86. Hu Q, Hao C, Tang S. From sepsis to acute respiratory distress syndrome (ARDS): Emerging preventive strategies based on molecular and genetic researches. Biosci Rep. 2020;40(5):1–9. Pravda J. Sepsis: Evidence-based pathogenesis and treatment. World J Crit Care Med. 2021;10(4):66–80. Vincent JL, Akça S, De Mendonça A, Haji-Michael P, Sprung C, Moreno R, et al. The epidemiology of acute respiratory failure in critically III patients. Chest. 2002;121(5):1602–9. Garland A, Dawson N V., Altmann I, Thomas CL, Phillips RS, Tsevat J, et al. Outcomes up to 5 years after severe, acute respiratory failure. Chest. 2004;126(6):1897–904. Dummer J, Stokes T. Improving continuity of care of patients with respiratory disease at hospital discharge. Breathe. 2020;16(3):1–8. Palakshappa JA, Krall JTW, Belfield LT, Files DC. Long-Term Outcomes in Acute Respiratory Distress Syndrome. Crit Care Clin. 2021 Oct;37(4):895–911. Rolfsen M, Mart MF, Sevin CM, Kieffer H, Krasinski DJ, Ferrante LE, et al. Communication of Post Intensive Care Syndrome: What Providers Reportedly Do and What Patients Remember. Am J Respir Crit Care Med. 2025;211(Abstracts):A1172–A1172. Meservey AJ, Burton MC, Priest J, Teneback CC, Dixon AE. Risk of Readmission and Mortality Following Hospitalization with Hypercapnic Respiratory Failure. Lung. 2020;198(1):121–34. Adler D, Peṕin JL, Dupuis-Lozeron E, Espa-Cervena K, Merlet-Violet R, Muller H, et al. Comorbidities and subgroups of patients surviving severe acute hypercapnic respiratory failure in the intensive care unit. Am J Respir Crit Care Med. 2017;196(2):200–7. Boelens YFN, Melchers M, Van Zanten ARH. Poor physical recovery after critical illness: Incidence, features, risk factors, pathophysiology, and evidence-based therapies. Curr Opin Crit Care. 2022;28(4):409–16. Herridge MS, Moss M, Hough CL, Hopkins RO, Rice TW, Bienvenu OJ, et al. Recovery and outcomes after the acute respiratory distress syndrome (ARDS) in patients and their family caregivers. Intensive Care Med. 2016;42(5):725–38. Herridge MS, Tansey CM, Matté A, Tomlinson G, Diaz-Granados N, Cooper A, et al. Functional Disability 5 Years after Acute Respiratory Distress Syndrome. N Engl J Med. 2011 Apr 7;364(14):1293–304. Heesakkers H, Van Der Hoeven JG, Corsten S, Janssen I, Ewalds E, Simons KS, et al. Clinical Outcomes among Patients with 1-Year Survival Following Intensive Care Unit Treatment for COVID-19. Jama. 2022;327(6):559–65. Gerth AMJ, Hatch RA, Young JD, Watkinson PJ. Changes in health-related quality of life after discharge from an intensive care unit: a systematic review. Anaesthesia. 2019;74(1):100–8. Oeyen SG, Vandijck DM, Benoit DD, Annemans L, Decruyenaere JM. Quality of life after intensive care: A systematic review of the literature. Crit Care Med. 2010;38(12):2386–400. Watson RS, Asaro LA, Hutchins L, Bysani GK, Killien EY, Angus DC, et al. Risk factors for functional decline and impaired quality of life after pediatric respiratory failure. Am J Respir Crit Care Med. 2019;200(7):900–9. Mayer KP, Welle MM, Evans CG, Greenhill BG, Montgomery-Yates AA, Dupont-Versteegden EE, et al. Muscle Power is Related to Physical Function in Patients Surviving Acute Respiratory Failure: A Prospective Observational Study. Am J Med Sci. 2021;361(3):310–8. Welch SA, Ward RE, Beauchamp MK, Leveille SG, Travison T, Bean JF. The Short Physical Performance Battery (SPPB): A Quick and Useful Tool for Fall Risk Stratification Among Older Primary Care Patients. J Am Med Dir Assoc. 2021;22(8):1646–51. Pavasini R, Guralnik J, Brown JC, di Bari M, Cesari M, Landi F, et al. Short Physical Performance Battery and all-cause mortality: Systematic review and meta-analysis. BMC Med. 2016;14(1):1–9. Western MJ, Malkowski OS. Associations of the Short Physical Performance Battery (SPPB) with Adverse Health Outcomes in Older Adults: A 14-Year Follow-Up from the English Longitudinal Study of Ageing (ELSA). Int J Environ Res Public Health. 2022;19(23). Voiriot G, Oualha M, Pierre A, Salmon-Gandonnière C, Gaudet A, Jouan Y, et al. Chronic critical illness and post-intensive care syndrome: from pathophysiology to clinical challenges. Ann Intensive Care. 2022;12(1). Intensive Care Unit. Intensive Care 2020 and beyond : Co-developing the future. 2020. 6–15 p. Turnbull AE, Lee EM, Dinglas VD, Beesley S, Bose S, Banner-Goodspeed V, et al. Health Expectations and Quality of Life After Acute Respiratory Failure: A Multicenter Prospective Cohort Study. Chest. 2023;164(1):114–23. Snowden S, Dahlén SE, Wheelock CE. Application of metabolomics approaches to the study of respiratory diseases. Bioanalysis. 2012;4(18):2265–90. Stringer KA, McKay RT, Karnovsky A, Quémerais B, Lacy P. Metabolomics and its application to acute lung diseases. Front Immunol. 2016;7(FEB). Patti GJ, Yanes O, Siuzdak G. Metabolomics: the apogee of the omic triology. Nat Rev Mol Cell Biol. 2012;13(4):2504. Langley RJ, Tsalik EL, Van Velkinburgh JC, Glickman SW, Rice BJ, Wang C, et al. Sepsis: An integrated clinico-metabolomic model improves prediction of death in sepsis. Sci Transl Med. 2013;5(195):1–18. Langley RJ, Migaud ME, Flores L, Thompson JW, Kean EA, Mostellar MM, et al. A metabolomic endotype of bioenergetic dysfunction predicts mortality in critically ill patients with acute respiratory failure. Sci Rep. 2021;11(1):1–12. Rogers AJ, McGeachie M, Baron RM, Gazourian L, Haspel JA, Nakahira K, et al. Metabolomic derangements are associated with mortality in critically ill adult patients. PLoS One. 2014;9(1):1–7. Langley RJ, Tipper JL, Bruse S, Baron RM, Tsalik EL, Huntley J, et al. Integrative “omic” analysis of experimental bacteremia identifies a metabolic signature that distinguishes human sepsis from systemic inflammatory response syndromes. Am J Respir Crit Care Med. 2014;190(4):445–55. Lelubre C, Vincent JL. Mechanisms and treatment of organ failure in sepsis. Nat Rev Nephrol. 2018;14(7):417–27. Tsalik EL, Willig LK, Rice BJ, van Velkinburgh JC, Mohney RP, McDunn JE, et al. Renal systems biology of patients with systemic inflammatory response syndrome. Kidney Int. 2015 Oct;88(4):804–14. Sorriento D, Di Vaia E, Iaccarino G. Physical Exercise: A Novel Tool to Protect Mitochondrial Health. Front Physiol. 2021;12(April):1–14. Brand MD, Orr AL, Perevoshchikova I V., Quinlan CL. The role of mitochondrial function and cellular bioenergetics in ageing and disease. Br J Dermatol. 2013;169(SUPPL.2):1–8. Morris PE, Berry MJ, Files DC, Thompson JC, Hauser J, Flores L, et al. Standardized rehabilitation and hospital length of stay among patients with acute respiratory failure a randomized clinical trial. JAMA - J Am Med Assoc. 2016;315(24):2694–702. Gandotra S, Lovato J, Case D, Bakhru RN, Gibbs K, Berry M, et al. Physical function trajectories in survivors of acute respiratory failure. Ann Am Thorac Soc. 2019;16(4):471–7. Faquih T, van Smeden M, Luo J, Le Cessie S, Kastenmüller G, Krumsiek J, et al. A workflow for missing values imputation of untargeted metabolomics data. Metabolites. 2020;10(12):1–23. Dekermanjian JP, Shaddox E, Nandy D, Ghosh D, Kechris K. Mechanism-aware imputation: a two-step approach in handling missing values in metabolomics. BMC Bioinformatics. 2022;23(1):1–17. Schiffman C, Petrick L, Perttula K, Yano Y, Carlsson H, Whitehead T, et al. Filtering procedures for untargeted lc-ms metabolomics data. BMC Bioinformatics. 2019;20(1):1–10. Bourgon R, Gentleman R, Huber W. Independent filtering increases detection power for high-throughput experiments. Proc Natl Acad Sci U S A. 2010;107(21):9546–51. RNABioUSA/arfqol-metabolomics [Internet]. [cited 2025 Jul 24]. Available from: https://github.com/RNABioUSA/arfqol-metabolomics Barupal DK, Fiehn O. Chemical Similarity Enrichment Analysis (ChemRICH) as alternative to biochemical pathway mapping for metabolomic datasets. Sci Rep. 2017;7(1):1–11. Pang Z, Lu Y, Zhou G, Hui F, Xu L, Viau C, et al. MetaboAnalyst 6.0: towards a unified platform for metabolomics data processing, analysis and interpretation. Nucleic Acids Res. 2024;52(W1):W398–406. Ewald JD, Zhou G, Lu Y, Kolic J, Ellis C, Johnson JD, et al. Web-based multi-omics integration using the Analyst software suite. Nature Protocols. Springer US; 2024. Liaw A, Wiener M. The R Journal: Classification and regression by randomForest. R J. 2002;2(3):18–22. Breiman L. Random Forests. Mach Learn. 2001 Oct;45(1):5–32. Ben G, Jonah G, Imad A, Sam B. rstanarm: {Bayesian} applied regression modeling via {Stan} [Internet]. 2020 [cited 2025 Aug 4]. Available from: https://mc-stan.org/rstanarm Vehtari A, Gelman A, Gabry J. Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. Stat Comput. 2017;27(5):1413–32. Membrez M, Migliavacca E, Christen S, Yaku K, Trieu J, Lee AK, et al. Trigonelline is an NAD+ precursor that improves muscle function during ageing and is reduced in human sarcopenia. Nat Metab. 2024;6(3):433–47. Traish AM, Abdallah B, Traish AM, Yu G, Traish AM. Androgen deficiency and mitochondrial dysfunction: Implications for fatigue., muscle dysfunction., insulin resistance., diabetes, and cardiovascular disease. Horm Mol Biol Clin Investig. 2011;8(1):431–44. Ahmad I, Newell-Fugate AE. Role of androgens and androgen receptor in control of mitochondrial function. Am J Physiol Cell Physiol. 2022;323(3):C835–46. Holeček M. Branched-chain amino acids in health and disease: metabolism, alterations in blood plasma, and as supplements. Nutr Metab (Lond). 2018 Dec 3;15(1):33. Pinto-Plata V, Casanova C, Divo M, Tesfaigzi Y, Calhoun V, Sui J, et al. Plasma metabolomics and clinical predictors of survival differences in COPD patients. Respir Res. 2019;20(1):1–8. Lovell D, Miller D, Capra J, Bradley A. Never mind the metrics -- what about the uncertainty? Visualising confusion matrix metric distributions. 2022; Jones GL, Qin Q. Markov Chain Monte Carlo in Practice. Annu Rev Stat Its Appl. 2022;9:557–78. Lauwers C, De Bruyn L, Langouche L. Impact of critical illness on cholesterol and fatty acids: insights into pathophysiology and therapeutic targets. Intensive Care Med Exp . 2023;11(1). Oh JH, Chae G, Song JW. Blood lipid profiles as a prognostic biomarker in idiopathic pulmonary fibrosis. Respir Res. 2024;25(1):1–9. Zhang D, Zhu Y, Su Y, Yu M, Xu X, Wang C, et al. Taurochenodeoxycholic acid inhibits the proliferation and invasion of gastric cancer and induces its apoptosis. J Food Biochem. 2022;46(3):1–9. Bao L, Hao D, Wang X, He X, Mao W, Li P. Transcriptome investigation of anti‐inflammation and immuno‐regulation mechanism of taurochenodeoxycholic acid. BMC Pharmacol Toxicol. 2021;22(1):1–11. Slijepcevic D, Roscam Abbing RLP, Katafuchi T, Blank A, Donkers JM, van Hoppe S, et al. Hepatic uptake of conjugated bile acids is mediated by both sodium taurocholate cotransporting polypeptide and organic anion transporting polypeptides and modulated by intestinal sensing of plasma bile acid levels in mice. Hepatology. 2017;66(5):1631–43. Salhab A, Amer J, Lu Y, Safadi R. Sodium + /taurocholate cotransporting polypeptide as target therapy for liver fibrosis. Gut. 2022;71(7):1373–85. Nickler M, Ottiger M, Steuer C, Huber A, Anderson JB, Müller B, et al. Systematic review regarding metabolic profiling for improved pathophysiological understanding of disease and outcome prediction in respiratory infections. Respir Res. 2015;16(1). Amasene M, Besga A, Medrano M, Urquiza M, Rodriguez-Larrad A, Tobalina I, et al. Nutritional status and physical performance using handgrip and SPPB tests in hospitalized older adults. Clin Nutr. 2021;40(11):5547–55. Brydges CR, Fiehn O, Mayberg HS, Schreiber H, Dehkordi SM, Bhattacharyya S, et al. Indoxyl sulfate, a gut microbiome-derived uremic toxin, is associated with psychic anxiety and its functional magnetic resonance imaging-based neurologic signature. Sci Rep. 2021;11(1):1–14. Weber D, Oefner PJ, Hiergeist A, Koestler J, Gessner A, Weber M, et al. Low urinary indoxyl sulfate levels early after transplantation reflect a disrupted microbiome and are associated with poor outcome. Blood. 2015 Oct 1;126(14):1723–8. van der Spek A, Stewart ID, Kühnel B, Pietzner M, Alshehri T, Gauß F, et al. Circulating metabolites modulated by diet are associated with depression. Mol Psychiatry. 2023;28(9):3874–87. Chakrabarty G, Das S, Dhara P. Impact of nutritional therapy on outcomes in critical care : A review of guidelines and clinical evidence. 2025;7(1):81–5. Sbaih N, Hawthorne K, Lutes J, Cavallazzi R. Nutrition Therapy in Non-intubated Patients with Acute Respiratory Failure. Curr Nutr Rep. 2021;10(4):307–16. Derave W, Özdemir MS, Harris RC, Pottier A, Reyngoudt H, Koppo K, et al. β-Alanine supplementation augments muscle carnosine content and attenuates fatigue during repeated isokinetic contraction bouts in trained sprinters. J Appl Physiol. 2007 Nov;103(5):1736–43. Harris RC, Tallon MJ, Dunnett M, Boobis L, Coakley J, Kim HJ, et al. The absorption of orally supplied β-alanine and its effect on muscle carnosine synthesis in human vastus lateralis. Amino Acids. 2006;30(3 SPEC. ISS.):279–89. Schnuck JK, Sunderland KL, Kuennen MR, Vaughan RA. Characterization of the metabolic effect of β-alanine on markers of oxidative metabolism and mitochondrial biogenesis in skeletal muscle. J Exerc Nutr Biochem. 2016;20(2):34–41. Cesak O, Vostalova J, Vidlar A, Bastlova P, Student V. Carnosine and Beta-Alanine Supplementation in Human Medicine: Narrative Review and Critical Assessment. Nutrients. 2023;15(7):1–21. Saunders B, Elliott-Sale K, Artioli GG, Swinton PA, Dolan E, Roschel H, et al. β-Alanine supplementation to improve exercise capacity and performance: A systematic review and meta-Analysis. Br J Sports Med. 2017;51(8):658–69. Maté-Muñoz JL, Lougedo JH, Garnacho-Castaño M V., Veiga-Herreros P, Lozano-Estevan M del C, García-Fernández P, et al. Effects of β-alanine supplementation during a 5-week strength training program: a randomized, controlled study. J Int Soc Sports Nutr. 2018;15(1):1–12. Lo A, Chernoff H, Zheng T, Lo SH. Why significant variables aren’t automatically good predictors. Proc Natl Acad Sci U S A. 2015;112(45):13892–7. Galal A, Talal M, Moustafa A. Applications of machine learning in metabolomics: Disease modeling and classification. Front Genet. 2022;13(November):1–25. Griffiths WJ, Koal T, Wang Y, Kohl M, Enot DP, Deigner HP. Targeted metabolomics for biomarker discovery. Angew Chemie - Int Ed. 2010;49(32):5426–45. Additional Declarations Competing interest reported. We developed a predictive model using a Bayesian framework that incorporates seven key metabolite biomarkers to estimate future physical function in critically ill patients following hospital discharge. To support future clinical translation, a provisional patent application has been filed to secure intellectual property rights for this biomarker-based prediction method. Supplementary Files Additionalfile1.pptx File Name - Additional file 1 File format - Power point (pptx) Title of data - Figure S1 to S7 Description of data – The file contains the supplementary figures cited in the manuscript. Each slide of the power point contains one panel of supplementary figure. Additinalfile2.xlsx File name - Additional file 2 File format - excel Title of data - Table S1: Evaluation Metrics for Standard and Bayesian Logistic Regression Models Description of the data - Legend for table: Performance metrics of Standard Logistic Regression and Bayesian Logistic Regression models for predicting six-month physical function outcome (SPPB: “Good” vs. “Poor”) using discharge serum metabolites. When trained on the full dataset, both models achieved identical AUC (0.94) and accuracy (0.86). Under 5-fold cross-validation, the Bayesian model consistently outperformed the standard model across all evaluation metrics, including a higher aggregated AUC (0.88 vs. 0.86), accuracy (0.81 vs. 0.75), precision, recall, F1 score, MCC, and Cohen’s Kappa. Additinalfile3.xlsx File name - Additional file 3 File Format - excel Title of data - Metabolomics data and Patients Demographics Description of data - This dataset comprises semi‑quantitative raw metabolite measurements collected at three clinical time points for each volunteer. It also includes patient demographic information, comorbidities, other clinical assessment scores and SPPB scores recorded at each of those points. The file also has spreadsheet results for PLS-DA and pathway analysis. Cite Share Download PDF Status: Published Journal Publication published 21 Nov, 2025 Read the published version in Respiratory Research → Version 1 posted Editorial decision: Accepted 05 Nov, 2025 Reviews received at journal 03 Nov, 2025 Reviewers agreed at journal 31 Oct, 2025 Reviews received at journal 31 Oct, 2025 Reviewers agreed at journal 31 Oct, 2025 Reviewers invited by journal 30 Oct, 2025 Submission checks completed at journal 30 Oct, 2025 First submitted to journal 27 Oct, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7394034","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":538131942,"identity":"ceb9014b-a052-42ec-95a2-244c42be619f","order_by":0,"name":"Adeyeye I. Haastrup","email":"","orcid":"","institution":"University of South Alabama Frederick P. Whiddon College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Adeyeye","middleName":"I.","lastName":"Haastrup","suffix":""},{"id":538131943,"identity":"90ee0095-38c1-487b-a96c-6da989e5f62d","order_by":1,"name":"Justin T. Roberts","email":"","orcid":"","institution":"University of South Alabama Frederick P. Whiddon College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Justin","middleName":"T.","lastName":"Roberts","suffix":""},{"id":538131944,"identity":"ef703fe5-803d-4d59-81e7-c3a7ea2a171f","order_by":2,"name":"Sheetal Gandotra","email":"","orcid":"","institution":"University of Alabama-Birmingham College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Sheetal","middleName":"","lastName":"Gandotra","suffix":""},{"id":538131945,"identity":"69f2fdc0-e7da-481c-b0ed-b431d0bc8b0e","order_by":3,"name":"Emily M. Hartsell","email":"","orcid":"","institution":"University of South Alabama Frederick P. Whiddon College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Emily","middleName":"M.","lastName":"Hartsell","suffix":""},{"id":538131946,"identity":"1763ae81-76ea-4146-94b3-995571b4d90a","order_by":4,"name":"Grant T. Daly","email":"","orcid":"","institution":"University of South Alabama Frederick P. Whiddon College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Grant","middleName":"T.","lastName":"Daly","suffix":""},{"id":538131947,"identity":"326dc315-d029-417b-a84b-588dba0f26ca","order_by":5,"name":"Viktor M. Pastukh","email":"","orcid":"","institution":"University of South Alabama Frederick P. Whiddon College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Viktor","middleName":"M.","lastName":"Pastukh","suffix":""},{"id":538131948,"identity":"7ade7c1f-f205-401a-b59b-c1acf237d74a","order_by":6,"name":"Lina D. Purcell","email":"","orcid":"","institution":"Wake Forest Baptist Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Lina","middleName":"D.","lastName":"Purcell","suffix":""},{"id":538131949,"identity":"b4c37236-bd5c-45e1-a675-d5ba73e7fd13","order_by":7,"name":"Ryan G. Benton","email":"","orcid":"","institution":"University of South Alabama School of Computing","correspondingAuthor":false,"prefix":"","firstName":"Ryan","middleName":"G.","lastName":"Benton","suffix":""},{"id":538131950,"identity":"6a4711a6-24df-4129-8bd4-3a8f5cd9be0c","order_by":8,"name":"D. Clark Files","email":"","orcid":"","institution":"Wake Forest Baptist Medical Center","correspondingAuthor":false,"prefix":"","firstName":"D.","middleName":"Clark","lastName":"Files","suffix":""},{"id":538131951,"identity":"7abaea1b-e511-4dfb-b9e2-f63028f75697","order_by":9,"name":"Troy Stevens","email":"","orcid":"","institution":"University of South Alabama Frederick P. Whiddon College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Troy","middleName":"","lastName":"Stevens","suffix":""},{"id":538131952,"identity":"acec2ade-e0d7-4afc-a3ca-315360d8f7c9","order_by":10,"name":"Mark N. Gillespie","email":"","orcid":"","institution":"University of South Alabama Frederick P. Whiddon College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Mark","middleName":"N.","lastName":"Gillespie","suffix":""},{"id":538131953,"identity":"a64e378b-59eb-445f-9bc3-cb6421fc7336","order_by":11,"name":"Peter E. Morris","email":"","orcid":"","institution":"University of Alabama-Birmingham College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Peter","middleName":"E.","lastName":"Morris","suffix":""},{"id":538131954,"identity":"509e349b-38ff-4b0c-8d28-ba81f99ec1ab","order_by":12,"name":"Raymond J. Langley","email":"data:image/png;base64,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","orcid":"","institution":"University of South Alabama Frederick P. Whiddon College of Medicine","correspondingAuthor":true,"prefix":"","firstName":"Raymond","middleName":"J.","lastName":"Langley","suffix":""}],"badges":[],"createdAt":"2025-08-17 18:53:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7394034/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7394034/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12931-025-03423-2","type":"published","date":"2025-11-21T15:57:08+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":94957086,"identity":"684dc6fa-758c-4099-8eac-dfe5d0699e61","added_by":"auto","created_at":"2025-11-02 14:55:42","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":154056,"visible":true,"origin":"","legend":"","description":"","filename":"ARFSurvivorsQualityofLifeManuscript.AH7.docx","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/9e64cfc6c77ca9ab2a90514a.docx"},{"id":94957083,"identity":"50566bfa-22b5-476c-a651-364eaadc0bbf","added_by":"auto","created_at":"2025-11-02 14:55:42","extension":"json","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13768,"visible":true,"origin":"","legend":"","description":"","filename":"10df4d96545344eebd2e2091dcd841d6.json","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/49101a9d44dc4cd9248e1511.json"},{"id":94957087,"identity":"915d0e6b-b4dd-423a-9924-3132354358e4","added_by":"auto","created_at":"2025-11-02 14:55:42","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":160912,"visible":true,"origin":"","legend":"","description":"","filename":"ARFSurvivorsQualityofLifeManuscript.AH7TrackChanges.docx","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/d540c8bd1f75dfb99a652c3c.docx"},{"id":94987774,"identity":"ca081d5e-4b37-4385-92f4-2f73447c9bc8","added_by":"auto","created_at":"2025-11-03 07:02:31","extension":"xlsx","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9305,"visible":true,"origin":"","legend":"","description":"","filename":"Additinalfile2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/469b85387ec11659ff6f7487.xlsx"},{"id":94988889,"identity":"1dfbbe15-fb0c-4de5-a11a-c9ab86c1d41f","added_by":"auto","created_at":"2025-11-03 07:11:17","extension":"xlsx","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3510662,"visible":true,"origin":"","legend":"","description":"","filename":"Additinalfile3.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/8d0626d97d894bfe0cb8db1f.xlsx"},{"id":94957103,"identity":"fc6203c3-bc7b-401e-b949-0d85825df8cd","added_by":"auto","created_at":"2025-11-02 14:55:43","extension":"pptx","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8664789,"visible":true,"origin":"","legend":"","description":"","filename":"Additionalfile1.pptx","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/66f9885abbcce901844e6399.pptx"},{"id":94957104,"identity":"5994aed7-ca93-4c79-983f-94d87af6229d","added_by":"auto","created_at":"2025-11-02 14:55:43","extension":"pptx","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3887024,"visible":true,"origin":"","legend":"","description":"","filename":"MainFiguresOned1.pptx","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/4dd8bbf9d982fb34039eb044.pptx"},{"id":94957097,"identity":"70939ee2-04f6-44f5-8a2a-0fdd4699a30e","added_by":"auto","created_at":"2025-11-02 14:55:42","extension":"xml","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":115825,"visible":true,"origin":"","legend":"","description":"","filename":"10df4d96545344eebd2e2091dcd841d61enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/ef72290b3aeaa143d6e19392.xml"},{"id":94957095,"identity":"677a6382-7448-4901-8f91-39f0c1021bb9","added_by":"auto","created_at":"2025-11-02 14:55:42","extension":"pptx","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3887024,"visible":true,"origin":"","legend":"","description":"","filename":"MainFiguresOne.pptx","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/804fd777dca54f156c981322.pptx"},{"id":94988325,"identity":"bdc64a85-5e56-4604-b0aa-13e1516eb03d","added_by":"auto","created_at":"2025-11-03 07:08:38","extension":"jpeg","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1074,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/d5fb9830930233f0568f0317.jpeg"},{"id":94989127,"identity":"c7d1a4c9-19d5-4ad5-b498-4c788ac7aea2","added_by":"auto","created_at":"2025-11-03 07:12:13","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":935,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/ae60b876abe8a8935a1769a7.png"},{"id":94957102,"identity":"bcf56578-1bdb-45ee-bbf1-101f198c1a37","added_by":"auto","created_at":"2025-11-02 14:55:43","extension":"xml","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":114847,"visible":true,"origin":"","legend":"","description":"","filename":"10df4d96545344eebd2e2091dcd841d61structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/45de14f4bfea09b33f8a9013.xml"},{"id":94957100,"identity":"1cd4650b-e643-4c3f-a9f8-0fb68c3ee1eb","added_by":"auto","created_at":"2025-11-02 14:55:42","extension":"html","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":123659,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/ac3c6c0da3c88b40e87c025f.html"},{"id":94988434,"identity":"94f226a8-74b0-4488-922e-e5ee4ba742a8","added_by":"auto","created_at":"2025-11-03 07:09:02","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":175097,"visible":true,"origin":"","legend":"\u003cp\u003eTitle: Volcano Plots Depicting Differential Serum Metabolites at Three Clinical Time Points. Legend: Volcano plots of metabolomic data comparing differential abundance of metabolites (p-value \u0026lt; 0.05) at three clinical time points among ARF survivors: (A) day 1 (ICU admission), 21 differential metabolites; (B) day 5 (post-admission), 21 differential metabolites; (C) at hospital discharge, 67 differential metabolites. Red points indicate upregulated differential metabolites, and purple points indicate those downregulated in participants with poor SPPB scores compared to those with good SPPB scores. Statistical threshold is marked by a dashed horizontal line (Wilcoxon rank sum p\u0026lt;0.05).\u003c/p\u003e","description":"","filename":"MainFiguresOne1.png","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/116e37375ef0a69aba27c70c.png"},{"id":94988694,"identity":"1223fdf4-cf1f-43f9-9cd2-e83eb2e7f01c","added_by":"auto","created_at":"2025-11-03 07:10:24","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":72718,"visible":true,"origin":"","legend":"\u003cp\u003eTitle: Enriched Metabolic Pathways in ARF Survivors Across Three Clinical Time Points from ICU Admission to Discharge. Legend: Pathway enrichment based on differential metabolites measured at three clinical time points: (A) ICU admission (day 1), (B) day 5 post-admission, and (C) hospital discharge. Circle size corresponds to the relative impact of each metabolic pathway, and circle color intensity (yellow to red) reflects statistical significance (−log10(p-value)).\u003c/p\u003e","description":"","filename":"MainFiguresOne2.png","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/0a40503a3b382bdf13ab389e.png"},{"id":94957092,"identity":"41edc773-0b2e-46cf-9d64-bbd243328c5f","added_by":"auto","created_at":"2025-11-02 14:55:42","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":75085,"visible":true,"origin":"","legend":"\u003cp\u003eTitle: Bayesian Logistic Regression Model Classifying ARF Survivors Based on Selected Serum Metabolites at Discharge. Legend: Figure 3. Bayesian logistic regression model for classification of SPPB outcome (“Good” vs. “Poor”) at discharge using a seven-metabolite panel. Normal (0, 2.5) priors were placed on the intercept and all coefficients to balance regularization with data-driven inference. (A) Posterior distributions of each metabolite’s log-odds coefficient (vertical lines = medians; shaded regions = 95% credible intervals). Positive values indicate increased odds of a “Good” SPPB score; negative values indicate decreased odds. (B) Confusion matrix highlighting the Bayesian model's prediction accuracy (C) ROC curve demonstrating robust predictive performance (AUC = 0.94).\u003c/p\u003e","description":"","filename":"MainFiguresOne3.png","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/438bfe453e4d301979f3907b.png"},{"id":94988608,"identity":"cf0d1f83-f910-4787-b912-facbc2cde049","added_by":"auto","created_at":"2025-11-03 07:10:05","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":63040,"visible":true,"origin":"","legend":"\u003cp\u003eTitle: Cross Validation of Bayesian Model Classifying ARF Survivors. Legend: Performance evaluation of the Bayesian logistic regression model predicting SPPB outcomes (\"Good\" vs. \"Poor\") based on serum metabolites measured at hospital discharge. (A) Confusion matrix and (B) Receiver Operating Characteristic (ROC) curve illustrating the performance from 5-fold cross-validation (AUC = 0.88). (C) Confusion matrix and (D) ROC curve derived from 100 model evaluations (repeated [20×5-fold] cross-validation (AUC = 0.88).\u003c/p\u003e","description":"","filename":"MainFiguresOne4.png","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/fcf35bfde3fb6666e3f033fb.png"},{"id":94957099,"identity":"575fab2a-e4f2-4309-b50a-fc98943904ff","added_by":"auto","created_at":"2025-11-02 14:55:42","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":112641,"visible":true,"origin":"","legend":"\u003cp\u003eTitle: Line Plot of SPPB Predictive Metabolites and Beta-Alanine. Legend: Longitudinal changes in selected serum metabolites across three clinical time points, day 1 (ICU admission), day 5 post-admission, and hospital discharge, stratified by six-month physical function (SPPB) groups. Blue lines and points denote patients with “Good” SPPB scores; red lines and points denote those with “Poor” SPPB scores. Data points are group means and error bars represent ± standard error of mean (SEM). Each panel shows one metabolite as labeled in the title, 7 metabolites predictive of six-month SPPB outcomes, and beta-alanine which has an established role in muscle metabolism.\u003c/p\u003e","description":"","filename":"MainFiguresOne5.png","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/b275e0435f1e7c28048393f9.png"},{"id":96650115,"identity":"209884ec-0345-4776-a3a5-ca05c4251ca4","added_by":"auto","created_at":"2025-11-24 16:08:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1064689,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/b214e09a-2b76-440a-bcb9-2f7c60bba80b.pdf"},{"id":94957091,"identity":"ffb26119-daa7-479e-b9c9-4184178a5c14","added_by":"auto","created_at":"2025-11-02 14:55:42","extension":"pptx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":8664789,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFile Name - Additional file 1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFile format - Power point (pptx)\u003c/p\u003e\n\u003cp\u003eTitle of data - Figure S1 to S7\u003c/p\u003e\n\u003cp\u003eDescription of data – The file contains the supplementary figures cited in the manuscript. Each slide of the power point contains one panel of supplementary figure.\u003c/p\u003e","description":"","filename":"Additionalfile1.pptx","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/278ed29e6d77b49143f9d0e7.pptx"},{"id":94957088,"identity":"2d595900-0b5f-433e-9ce4-c1b153e9bd0b","added_by":"auto","created_at":"2025-11-02 14:55:42","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":9305,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFile name - Additional file 2\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFile format - excel\u003c/p\u003e\n\u003cp\u003eTitle of data - Table S1: Evaluation Metrics for Standard and Bayesian Logistic Regression Models\u003c/p\u003e\n\u003cp\u003eDescription of the data - Legend for table: Performance metrics of Standard Logistic Regression and Bayesian Logistic Regression models for predicting six-month physical function outcome (SPPB: “Good” vs. “Poor”) using discharge serum metabolites. When trained on the full dataset, both models achieved identical AUC (0.94) and accuracy (0.86). Under 5-fold cross-validation, the Bayesian model consistently outperformed the standard model across all evaluation metrics, including a higher aggregated AUC (0.88 vs. 0.86), accuracy (0.81 vs. 0.75), precision, recall, F1 score, MCC, and Cohen’s Kappa.\u003c/p\u003e","description":"","filename":"Additinalfile2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/2c7faa6931b2d62b0b315643.xlsx"},{"id":94957093,"identity":"e30e8412-9292-4c25-a5d2-558314adb637","added_by":"auto","created_at":"2025-11-02 14:55:42","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":3510662,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFile name - Additional file 3\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFile Format - excel\u003c/p\u003e\n\u003cp\u003eTitle of data - Metabolomics data and Patients Demographics\u003c/p\u003e\n\u003cp\u003eDescription of data - This dataset comprises semi‑quantitative raw metabolite measurements collected at three clinical time points for each volunteer. It also includes patient demographic information, comorbidities, other clinical assessment scores and SPPB scores recorded at each of those points. The file also has spreadsheet results for PLS-DA and pathway analysis.\u003c/p\u003e","description":"","filename":"Additinalfile3.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7394034/v1/f0bcaf98212ccd03006b05cf.xlsx"}],"financialInterests":"Competing interest reported. We developed a predictive model using a Bayesian framework that incorporates seven key metabolite biomarkers to estimate future physical function in critically ill patients following hospital discharge. To support future clinical translation, a provisional patent application has been filed to secure intellectual property rights for this biomarker-based prediction method.","formattedTitle":"Metabolomic Biomarkers Predict Long-term Physical Function in Survivors of Acute Respiratory Failure","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAcute respiratory failure (ARF) is a life-threatening condition characterized by inadequate blood oxygenation with or without hypercapnia [1,2]. It is often associated with critical conditions such as sepsis, chronic obstructive pulmonary disease (COPD), and acute respiratory distress syndrome (ARDS), and imposes a substantial socioeconomic burden on affected patients and society as a whole [3,4]. A multinational study revealed that more than 50% of critically ill patients in the ICU have or will develop ARF, which places them at a 40% risk of mortality [5].\u003c/p\u003e\u003cp\u003eSurvivors of ARF often face multiple physical and cognitive difficulties that are collectively termed post-intensive care syndrome (PICS) after ICU discharge [6\u0026ndash;9]. About 46% of ARF survivors are at high risk of readmission or dying within 3\u0026ndash;4 months following initial ICU discharge which may be due to PICS-related disabilities [10,11]. Persistent physical disabilities after discharge from the ICU [6,12,13], reflecting impairments in muscle strength, mobility, and overall function long after the acute illness [14], are core components of PICs. In survivors of COVID-19-related ARDS, the incidence of physical function impairment is approximately 74% at 1 year after ICU treatment [15]. The quality of life decrement for these patients is influenced significantly by various factors, including underlying health conditions, demographics, ICU treatment, and the medical, pharmacological, and rehabilitative follow-up treatments [16\u0026ndash;19].\u003c/p\u003e\u003cp\u003eThe Short Physical Performance Battery (SPPB) is a tool commonly used to measure physical performance in the elderly [20]. It has been shown to correlate with various important health outcomes, including the ability to perform daily activities, risk of hospital admission or readmission, overall functional status, and mortality rates in critically ill and elderly subjects [21,22]. While the etiology of ARF is reasonably well understood, pathophysiologic mechanisms contributing to persistent poor physical performance in post-ICU patients, as well as reliable tools to predict long-term functional outcomes remain elusive [12,23]. The long-term quality of life outcomes for ARF survivors, identified as a key research priority by critical care professional societies and pulmonary physicians, emphasizes the urgent need for research focused on delineating the pathophysiologic changes that occur at the molecular level [24,25].\u003c/p\u003e\u003cp\u003eMetabolomics is increasingly utilized to profile interactions at a system level in patients with critical illness [26,27]. This approach may enable a comprehensive and holistic understanding of the disease process as well as improvements in diagnosis, assessing response to therapy, monitoring disease progression, pathological mechanisms, biomarker, and pharmacologic target discovery [28].\u003c/p\u003e\u003cp\u003eOur previous studies identified bioenergetic dysregulation as a key factor in non-survivors of sepsis, ARF, and acute kidney injury (AKI) [29\u0026ndash;34]. The prominently affected pathways in non-survivors were bioenergetic, specifically NAD⁺ metabolism and β-oxidation, which are essential for mitochondrial bioenergetics, and these were accompanied by disruptions in acetylcarnitine, glutathione, bile acid, steroid, and fatty acid metabolism[30]. Building on this, because physical exercise performance is inherently linked to mitochondrial bioenergetics [35,36], we postulated that the dysregulation in bioenergetic pathways may follow a continuum, with varying degrees of severity correlating with a range of outcomes. For example, those with the most pronounced metabolic disruptions did not survive, while survivors with poor outcomes experience milder but still impactful bioenergetic dysregulation, that potentially impairs physical performance. In this study, we hypothesized that poor physical function in ARF survivors is associated with, and predicted by, bioenergetic dysfunction, as reflected by metabolomic and mitochondrial biomarkers.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eThis retrospective study involved ARF patients enrolled in the Standardized Rehabilitation for ICU Patients with Acute Respiratory Failure clinical trial at Wake Forest Baptist Medical Center in North Carolina (ClinicalTrials.gov Identifier, NCT00976833; registration date, 2009-09-11). Institutional review board approval was obtained, and informed consent was provided by patients or their legal representatives. ICU patients were enrolled based on a clinical diagnosis of ARF made by the attending physicians, following the inclusion and exclusion criteria previously outlined [30,37]. All patients were treated according to standard of care practices, comprising a variety of pharmacological therapies. To ensure comparability between the groups, the patients were matched for age, race, and sex. Demographics of the selected patients are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Global metabolomics was performed on patients' serum samples at three time points: ICU admission (n\u0026thinsp;=\u0026thinsp;70), five days post-admission (n\u0026thinsp;=\u0026thinsp;40), and ICU discharge (n\u0026thinsp;=\u0026thinsp;69). Before blood collection from a central line, the catheter was flushed with sterile saline and a waste sample was drawn to remove residual fluids or anticoagulants, ensuring uncontaminated samples. Samples were collected at enrollment, day 5 post-enrollment, and at ICU discharge. The smaller number of samples at day 5 and at discharge reflects both study design and patient availability. We anticipated that this timepoint would provide limited clinical utility beyond showing temporal trends, so we prioritized enrollment and discharge samples to maximize predictive power. Furthermore, several patients in both good and poor outcomes were discharged prior to day 5. There was one sample at discharge that was not measured by Metabolon and therefore removed from the analysis. At discharge, and at two-, four-, and six-months following ICU admission, we assessed physical function using the SPPB, an objective measure that evaluates gait speed, balance, and lower extremity strength [38].\u003c/p\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eSerum Metabolomics\u003c/span\u003e: Extraction and quantification of serum metabolites were performed by Metabolon Inc [30]. Briefly, 100ul of each sample was prepared with recovery standards using the automated MicroLab STAR\u0026reg; system (Hamilton Company). Sample runs were performed on C18 and HILIC columns coupled to a Thermo Scientific Q-Exactive high-resolution/accurate mass spectrometer with the orbitrap mass analyzer operated at 35,000 mass resolution in negative and positive ionization modes. Metabolon's hardware and proprietary software were used for spectral data extraction and compound identification. Artifacts and background noise were removed. Peaks were quantified using the area-under-the-curve method. Specifically, identifications are based on a narrow retention index, mass accuracy within \u0026plusmn;\u0026thinsp;10 ppm, and MS/MS spectral matching to authentic standards (forward and reverse scores). The scaled intensity of metabolites was used to provide semi-quantitative metabolite values, wherein each value quantifies the relative concentration of a metabolite in a sample. The mass area was normalized to account for sample volume, blanks, quality controls, and internal standards. Variance filtering for repeatability (using QC samples) was performed using internal standards and endogenous metabolites, with relative standard deviation thresholds of 3% and 9%, respectively.\u003c/p\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eMetabolomic Data Processing and Statistical Comparisons\u003c/span\u003e: Metabolites with greater than 30% missing points were excluded from further processing [39,40]. Missing data points were imputed with mechanism-aware imputation algorithm [40]. A Shapiro-Wilk test performed on the data before and after imputation gave p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05, indicating non-normal distribution of the metabolite data. To improve signal-to-noise ratio and enhance downstream modeling, metabolites with a relative median absolute deviation less than 0.25, representing near-constant features with limited statistical power, were excluded [41,42]. Metabolite intensities were transformed to log scale in order reduce skewness. To explore differences in metabolite levels between functional outcome groups, we applied the Wilcoxon rank sum test (two-sided) to compare survivors with poor (SPPB\u0026thinsp;\u0026le;\u0026thinsp;6) and good (SPPB\u0026thinsp;\u0026ge;\u0026thinsp;7) scores. Metabolites with p\u0026thinsp;\u0026lt;\u0026thinsp;0.05 were considered to show the strongest differences and are hereafter referred to as \u0026ldquo;differential metabolites\u0026rdquo;. We explored false discovery rate (FDR) correction using multiple methods including the Benjamini-Yekutieli procedure; however, due to the highly discrete distribution of p-values, modest sample size and limited variability among test statistics, the adjustment yielded uniformly non-significant results. As such, FDR-adjusted results were not used for inference. Metabolomic profiles were analyzed separately at three timepoints: day 1, day 5 post admission, and at hospital discharge. Multivariate hierarchical clustering of differential metabolites was performed to assess temporal changes in metabolomic profiles between good and poor physical function groups across the three time points. The data were analyzed using custom-developed scripts in RStudio version 4.4.1 [43].\u003c/p\u003e\u003cp\u003eStatistical analysis of the demographic data was conducted using Wilcoxon rank-sum test for continuous variables and Fisher\u0026rsquo;s exact test for categorical variables. A p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered significant when comparing poor and good physical performance groups.\u003c/p\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eChemical Similarity Enrichment, Pathway and Meta-Analysis\u003c/span\u003e: Chemical similarity enrichment analysis was performed using ChemRICH, which enabled us to identify classes of compounds that were up- or downregulated. ChemRICH also offers the advantage of identifying clusters of chemically similar metabolites, revealing potential new metabolite sets that may be relevant [44]. Pathway analysis was performed using MetaboAnalyst 6.0 focusing on differential metabolites to identify specific dysregulated pathways in subjects with poor physical function [45,46]. A meta-analysis was conducted to evaluate the association of self-declared race (white or black) and gender (male or female) with differential metabolites that were either upregulated or downregulated. In addition, Fisher's Exact Test was used to assess the association between physical function outcomes, categorized as good vs. poor SPPB, and clinical parameters including APACHE III score, body mass index (BMI), mean arterial pressure (MAP), heart rate (pulse), respiratory rate, fraction of inspired oxygen (FiO₂), and arterial partial pressure of carbon dioxide (PaCO₂), as well as comorbidities present at admission.\u003c/p\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eMetabolite Feature Selection, Classification, and Model Validation\u003c/span\u003e: As indicated subsequently, the highest number of differential metabolites and dysregulated pathways was observed in the poor SPPB group at discharge. To streamline relevant features for sample classification, we applied a filtering process on differential metabolites measured at discharge. Log transformed data were mean-centered and scaled to unit variance. Using the publicly available online version of MetaboAnalyst\u0026trade; software, a Partial Least Squares Discriminant Analysis (PLS-DA) was performed to classify the patients into the respective SPPB categories, the classification being executed utilizing the biomarker candidates comprising only the differential metabolites. Additionally, a supervised machine learning, Random Forest classification model was implemented to rank biomarker candidates features based on their discriminatory power, using the randomForest and caret R packages [47,48]. Ten metabolites identified at this stage were subjected to Spearman correlation analysis, to eliminate redundant features exhibiting high collinearity or duplicative discriminatory effects. A correlation threshold of 0.8 was defined, and biomarker pairs exhibiting absolute correlation values exceeding this threshold were identified as highly correlated features. Furthermore, a Bayesian logistic regression model incorporating leave-one-out cross-validation was utilized (rstanarm and loo packages in R) to assess the impact of inclusion or exclusion of metabolites on model performance [49,50]. Based on this iterative evaluation, seven metabolite biomarkers identified as the most predictive of patients' physical function were selected for standard logistic regression analysis with 5-fold cross-validation. Given the strong performance of the standard model, we further evaluated the predictive power of these metabolites using a Bayesian logistic regression framework for comparison. To evaluate the generalizability of the model to unseen data, a 5-fold cross-validation procedure was conducted. To validate the robustness and stability of the model across variable data partitions, a repeated k-fold cross-validation procedure was implemented, through twenty repetitions of 5-fold cross-validation. The execution parameters included five independent Markov chains, each with 2,000 iterations and a fixed random seed of 12345 for reproducibility.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eTo determine if bioenergetic metabolites are predictive of six-month physical function in survivors of ARF, we performed broad spectrum metabolomic analysis of serum samples collected at enrollment (day 1), day 5 post enrollment, and patient discharge. Seventy patients were selected from the Standardized Rehabilitation for ICU Patients with Acute Respiratory Failure cohort. Patients were matched for age, race, and sex (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Half the patients were determined to have a \u0026ldquo;good\u0026rdquo; physical performance, (SPPB\u0026thinsp;\u0026ge;\u0026thinsp;7), while the other half had a \u0026ldquo;poor\u0026rdquo; physical performance (SPPB\u0026thinsp;\u0026le;\u0026thinsp;6). Demographics of the selected patients are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePatients Demographics\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVariable\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGood\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePoor\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eP-value\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003en-value\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e35\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e35\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSPPB score\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e10 [8.5\u0026ndash;10]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3 [0\u0026ndash;5]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAPACHE III score\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e70 [57.5\u0026ndash;82.5]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e69 [60.5\u0026ndash;79]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.764\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMAP (mmhg)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e65 [56\u0026ndash;93]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e68 [57\u0026ndash;100]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.477\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePulse (bpm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e103 [85\u0026ndash;123]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e106 [85.5\u0026ndash;114]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.577\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRespiratory rate (breaths/min)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e24 [20.5\u0026ndash;30.5]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e22 [19.5\u0026ndash;28.5]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.303\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePCO\u003csub\u003e2\u003c/sub\u003e (mmhg)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e44 [34-52.25]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e37.6 [32.5\u0026ndash;43.7]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.127\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePAO\u003csub\u003e2\u003c/sub\u003e (mmhg)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e95 [77\u0026ndash;121]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e91.6 [75.05\u0026ndash;112.5]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.703\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFIO\u003csub\u003e2\u003c/sub\u003e (mmhg)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.5 [0.4\u0026ndash;0.6]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.6 [0.4\u0026ndash;0.7]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.386\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBUN (mg/dl)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e19 [15\u0026ndash;29]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e18 [11.5\u0026ndash;28]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.577\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSerum albumin (g/dl)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3 [2.6\u0026ndash;3.3]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.7 [2.3\u0026ndash;3.2]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.127\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSerum bilirubin (mg/dl)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.8 [0.6\u0026ndash;1.05]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.9 [0.55\u0026ndash;1.15]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.991\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBlood glucose (mg/dl)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e137 [100\u0026ndash;204]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e132 [116.5-164.5]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.707\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBMI (kg/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e30.27 [26.6-36.33]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e29.67 [23.07\u0026ndash;34.35]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.280\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge (years)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e62 [54-69.5]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e59 [54-71.5]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.986\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLOS (days)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e9 [7-11.5]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e12 [8.5\u0026ndash;17]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.019\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSex\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e18 (51.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e18 (51.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e17 (48.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e17 (48.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRace\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBlack\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e11 (31.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e11 (31.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWhite\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e24 (68.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e24 (68.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNutrition\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e30 (85.7%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e29 (82.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e5 (14.3%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e6 (17.1%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHome oxygen\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e25 (71.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e25 (71.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e10 (28.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e10 (28.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDialysis\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e31 (88.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e33 (94.3%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e4 (11.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2 (5.7%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePhysical therapy\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e15 (42.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e15 (42.9%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e20 (57.1%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e20 (57.1%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eMass spectrometry analysis was performed by Metabolon Inc. as described in the methods. In total, 758 named biochemicals and 151 unnamed biochemicals were detected (total: 909 biochemicals). After removing near-constant features and restricting to named compounds, 742 biochemicals were retained for analysis. Samples identified as outliers by Metabolon Inc. were removed from the statistical analysis, reducing the number of samples to 66, 39, and 64 on day 1, day 5, and at discharge, respectively. Further analysis was performed on named biochemicals.\u003c/p\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eUnivariate and Multivariate Analysis Results\u003c/span\u003e: On day one, 21 serum metabolites concentrations were identified as having the strongest difference (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) between outcome groups. Patients with poorer physical outcomes exhibited a marked reduction in bioenergetic-related metabolites, including trigonelline, a metabolite of NAD⁺, which plays a key role in cellular energy metabolism (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA) [51].\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eAt day 5 post-admission, 21 serum metabolites differentiated patients with poor and good physical performance. Androgens play crucial roles in bioenergetics [52,53]. Amongst the differential metabolites, five androgen-related metabolites were lower in patients who had poor physical function (dehydroepiandrosterone sulfate [DHEA-S], androsterone sulfate, epiandrosterone sulfate, 5-alpha-androstan-3beta,17beta-diol disulfate, and androstenediol [3alpha, 17alpha] monosulfate) (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB).\u003c/p\u003e\u003cp\u003eRemarkably, 67 differential metabolites distinguished patients with poor versus good physical function at discharge. Among ARF survivors with good SPPB scores, 22 differential metabolites were decreased at discharge, compared to only 6 on day 5 and 4 on day 1. In contrast, survivors with poor SPPB scores showed 45 differential metabolite with decreased concentrations at discharge, 15 at day 5, and 17 at day 1. This highlights a noticeable shift in the metabolomic profile by the time of hospital discharge. Bioenergetic metabolites such as hydroxyl fatty acids, fructose, glycerides, alpha-ketoglutarate, and glycerol-3-phosphate were reduced in patients with poor physical function (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC).\u003c/p\u003e\u003cp\u003eAlthough the primary aim of this study was to identify predictive metabolite biomarkers and develop an FDR-independent predictive model, initial analyses revealed several metabolites with nominal significance (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05); however, these did not pass the Benjamini-Yakutieli correction (adjusted p\u0026thinsp;\u0026gt;\u0026thinsp;0.05).\u003c/p\u003e\u003cp\u003eMultivariate analysis of the metabolite trends between the good and poor SPPB score groups across time point showed some temporal differences reflecting dynamic metabolic shifts across the three time points (Additional file1: figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eChemical Similarity Enrichment and Pathway Analysis Results\u003c/span\u003e: We performed chemical and pathway enrichment analyses to evaluate chemical similarities among enriched metabolites, identify trends in metabolic pathway alterations during hospitalization, and highlight potential pathways that may be targeted for therapeutic intervention.\u003c/p\u003e\u003cp\u003eAnalysis of the metabolomic panel on day 1 reveals disruption in the catabolism of valine, leucine, and isoleucine in ARF survivors with poor SPPB scores (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). These branched-chain amino acids are essential for muscle metabolism and mitochondrial energy production [54]. Pathway analysis further underscores slight perturbations in cysteine and methionine metabolism. ChemRICH analysis highlights the decreased keto acids (branched-chain amino acid metabolites) and increased cholic acids (bile acids) which corresponds with metabolic pathways identified as impacted by the pathway analysis with MetaboAnalyst (Additional file 1: Figure \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e A).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eOn day five, pathway analysis demonstrates dysregulation in both primary bile acid biosynthesis and amino acid metabolism in group of ARF survivors with poor SPPB scores (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB). Similar to day one, ChemRICH analysis only revealed decreased levels of keto acids associated with group with poor SPPB scores (Additional file 1: Figure \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e B)\u003c/p\u003e\u003cp\u003eAt discharge, pathway analysis documented alterations in the tricarboxylic acid (TCA) cycle, suggesting disrupted cellular energy production with potential adverse effects on physical recovery in patients with poor SPPB scores. Alanine, aspartate and glutamate metabolism and glycerophospholipid metabolism are also key pathways dysregulated at discharge (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC).\u003c/p\u003e\u003cp\u003eBased on the ChemRICH analysis, carnitines emerged as the most elevated, while saturated fatty acids were the most decreased class of metabolites in patients with poor physical performance at the discharge time point (Additional file 1: Figure \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e C).\u003c/p\u003e\u003cp\u003eAlthough several pathways, including \"Valine, leucine, and isoleucine degradation\" and \"Primary bile acid biosynthesis,\" showed as dysregulated at earlier time points (ICU admission and day 5), their biological impact scores were limited, indicating subtle metabolic shifts. In contrast, pathway impact scores notably increased at discharge (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eMeta-Analysis\u003c/span\u003e: We were interested in determining if patient demographics (age, race, and sex) were associated with metabolomic changes. Overall, there was no association with age. On day 1, the only differential metabolite that was slightly downregulated in Black patients was 3-indoxyl sulfate.\u003c/p\u003e\u003cp\u003eFemales exhibited higher levels of the bile acid taurocholate at both day 1 and day 5 compared to males. 1-palmitoyl-2-palmitoleoyl-GPC, was also differentially elevated in White patients than in Black patients on day 5.\u003c/p\u003e\u003cp\u003eAt discharge, the metabolite profile showed no gender-based differential abundance. However, four metabolites were elevated in Black patients and three in White patients. Black patients showed higher levels of bile acids (taurochenodeoxycholate, hyocholate), threonate, and the peptide HWESASXX, which has been linked to worse outcomes in COPD [55]. In contrast to Black patients, White patients showed elevated levels of carnitine conjugates of medium and long chain fatty acids [(myristoleoylcarnitine (C14:1), palmitoleoylcarnitine (C16:1), 5-dodecenoylcarnitine (C12:1))], metabolites related to β-oxidation and mitochondrial function.\u003c/p\u003e\u003cp\u003eWe were also interested in the potential influence of co-existing illness on the metabolic profile of the ARF patients that may be related to poor or good physical function. No significant associations were observed between patient physical function and comorbidities, APACHE III scores, or other clinical measures such as mean arterial pressure (Additional file 1: Figure \u003cspan refid=\"MOESM3\" class=\"InternalRef\"\u003eS3\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003ePatient Classification Modeling\u003c/span\u003e: Given that the most differential serum metabolomic differences were observed at patient discharge compared to day 1 and day 5, we focused on this time point using the data of 30 participants with poor SPPB scores and 34 with good SPPB scores (Additional file 1: Figure \u003cspan refid=\"MOESM4\" class=\"InternalRef\"\u003eS4\u003c/span\u003eA \u0026ndash; S4C). We developed classification models to assess whether metabolomic profiles could anticipate the six-month SPPB scores. To identify the most relevant metabolites, we applied PLS-DA, Spearman correlation matrix, random forest feature selection, and evaluated using Bayesian leave-one-out cross-validation (Additional file 1: Figure \u003cspan refid=\"MOESM4\" class=\"InternalRef\"\u003eS4\u003c/span\u003e). Seven predictive metabolite biomarkers were selected for further analysis: cystine, glycerol 3-phosphate, hyocholate, 2-hydroxy-3-methylvalerate, taurocholenate sulfate, 1-oleoyl-2-docosahexaenoyl-GPC (18:1/22:6), and ximenoylcarnitine. These were subsequently used in a standard logistic regression model to assess their predictive performance. The standard logistic regression model achieved an accuracy of 0.86, along with a strong area under the curve (AUC) of 0.94. Following 5-fold cross-validation, the model continued to perform well, with an accuracy of 0.75 and an AUC of 0.86 (Additional file 1: Figure \u003cspan refid=\"MOESM5\" class=\"InternalRef\"\u003eS5\u003c/span\u003e). These results suggest good discriminative ability without evidence of overfitting and support its potential to generalize to independent validation cohorts.\u003c/p\u003e\u003cp\u003eTo compare with standard logistic regression, we assessed the predictive capacity of the seven metabolites using Bayesian logistic regression framework. The intercept\u0026rsquo;s posterior distribution of the posterior plot, centered near zero, suggests baseline equal odds of 'Poor' versus 'Good' physical condition when metabolites are average (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). A slightly positive posterior indicates a mild bias toward predicting 'Good' SPPB scores. The posterior distributions for cystine, glycerol 3-phosphate, and 2-hydroxy-3-methylvalerate lie predominantly above zero, indicating that higher levels of these metabolites credibly increase the odds of a \u0026lsquo;Good\u0026rsquo; SPPB score. In contrast, the posteriors for hyocholate, taurocholenate sulfate, 1-oleoyl-2-docosahexaenoyl-GPC, and ximenoylcarnitine lie below zero, suggesting that elevated concentrations are associated with poorer physical function. All metabolites present moderate posterior distributions with 95% credible intervals clear of zero, underscoring consistent associations in predicting physical function (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). The posterior predictive plot (Additional file 1: Figure S6 A) shows a close correspondence between simulated and observed outcomes, with the predictive intervals tightly enveloping the empirical data. This concordance indicates that the Bayesian model faithfully captures the underlying data structure and produces reliable predictions.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe confusion matrix and ROC curve of Bayesian model (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC) indicate strong model-fit for classifying patients into 'Good' and 'Poor' SPPB groups [56]. The trace plot indicates convergence to the target distribution, suggesting well-sampled posterior distributions (Additional file 1: Figure S6 B). Low autocorrelation in MCMC chains implies efficient Bayesian sampling, with reliable posterior estimates (Additional file 1: Figure S6 C) [57]. Patient classification under the Bayesian framework was further assessed using 5-fold and repeated k-fold cross-validation (100 model evaluations), achieving a robust AUC of 0.88 (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Collectively, these results underscore strong predictive associations between serum metabolites measured at hospital discharge and subsequent patient physical function. The consistently high AUC values and accuracy across multiple validation approaches indicate that the selected metabolite panel provides stable and effective predictive utility for future patient physical performance outcomes.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe line plot illustrates the trends of the seven predictive metabolites biomarkers and β-alanine at key time points during hospitalization, day 1 of ICU admission, day 5, and at discharge (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eBased on our prior work in ARF patients, we hypothesized that the most pronounced metabolomic differences between patients with good and poor physical function would occur at ICU admission, reflecting the acute severity of illness. However, our findings revealed a surprising pattern: despite clinical improvement by the time of discharge, the number of differential metabolites was threefold higher at discharge compared to ICU admission or mid-hospitalization (Additional file 1: Figure S7). The only metabolite that consistently differed at all three time points between patients with poor and good physical function was the bile acid taurochenodeoxycholate, showing higher mean and median values in those with poor SPPB test results. Additionally, bile acids were generally elevated across all time points in patients with poor physical performance, suggesting a consistent dysregulation of bile acid metabolism in this group. These findings are further supported by hierarchical clustering of multivariate metabolite profiles, which revealed some noticeable patterns between good and poor SPPB groups across time points, especially in differential metabolites related to bile acid and phosphatidylethanolamines metabolism and bioenergetics (Additional file 1: Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eChemRICH and pathway analysis (using MetaboAnalyst) identified similar, though slightly varied dysregulated metabolomic alterations at each time point. These findings suggest that while the overall metabolic disruptions are consistent, the specific pathways and metabolites involved may shift as the patient progresses through the phases of hospital care and recovery. Our previous research demonstrated a correlation between acute mortality in ARF patients and elevated carnitine levels alongside mitochondrial-related bioenergetic dysfunction [30]. Notably, ChemRICH analysis of metabolites at discharge revealed that carnitines constituted the most enriched biochemical set in patients with poor SPPB scores, supporting the association between carnitine upregulation and worse outcomes in critically ill patients (Additional file 1: Figure \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e C) [30]. Patients with poor physical function showed reduced lipid and fatty acid levels, and elevated bile acids (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC; Additional file 1: Figure \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e A, S2 C), consistent with prior studies linking low lipid levels in critical illness to worse outcomes [58,59]. These metabolomic differences suggest a shift in bile acid homeostasis, potentially reflecting altered hepatic metabolism, although serum albumin and bilirubin levels did not differ significantly between groups (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) [60\u0026ndash;63].\u003c/p\u003e\u003cp\u003eAt the discharge time point, disruption in glycerophospholipid metabolism in the poor SPPB group may impair membrane lipid turnover, which is critical for maintaining cellular integrity and membrane fluidity. Similar reductions in glycerophospholipid metabolism have been previously observed to correlate with poor outcomes in patients with COPD (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC and Additional file 1: Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e) [64].\u003c/p\u003e\u003cp\u003eNutritional status variations also could affect the metabolic disparities between patients with high and low SPPB scores [65]. The lower fatty acid and lipids levels observed with poor SPPB scores suggest that the increased bile acid production may be a compensatory metabolic response, potentially important for enhancing lipid digestion and absorption. On day 1, 3-indoxyl sulfate, a tryptophan-derived metabolite was downregulated in Black ARF survivors and has been implicated in microbial\u0026ndash;host interactions, gut homeostasis, and uremic toxicity [66,67]. Moreover, a diet modifiable metabolites, 1-palmitoyl-2-palmitoleoyl-GPC, was differentially elevated in White patients than in Black patients on day 5 [68]. Given that only few ARF patients in the cohort received nutritional therapy, and considering the known impact of nutritional status on both short- and long-term outcomes in critical illness, metabolomic profiling may offer valuable insights into abnormal metabolic signatures. These insights could inform personalized nutritional strategies to improve long-term functional outcomes in ARF patients [69,70].\u003c/p\u003e\u003cp\u003eIt is also noteworthy that the APACHE III score showed no significant association with poor physical performance. While this clinical metric provides an overall assessment of patient condition, it may not capture the nuanced biological variability that influences physical function in critically ill patients. Likewise, other important variables such as physical therapy, home oxygen use, and dialysis have no significant association with poor or good physical function. The lack of significant association in aforementioned parameters with the physical function status may be due to modest sample size used in this preliminary study. On the contrary, patients with longer ICU stays were significantly more likely to have poor physical function outcomes (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eWe observed decreased levels of β-alanine at discharge in patients with poor SPPB scores (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC). β-alanine is a central metabolite serving as a precursor in various key metabolic processes, including the synthesis of carnosine, a dipeptide of β -alanine and histidine, which supports muscle metabolism [71\u0026ndash;75]. β-alanine intake could thus potentially benefit ARF patients prone to poor physical performance and thus seem to warrant additional research into its clinical applications [71,72,74\u0026ndash;76].\u003c/p\u003e\u003cp\u003eThe strong performance of both logistic regression and Bayesian logistic regression models demonstrates that metabolomic profiling at the time of ICU discharge can effectively predict patients\u0026rsquo; subsequent physical performance (Additional file 2: Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). While frequentist logistic regression shares a similar foundational framework with its Bayesian counterpart, the Bayesian approach demonstrated superior performance during cross-validation, achieving an AUC of 0.88 compared to 0.86 with standard logistic regression. Moreover, the Bayesian model yielded fewer false positives and false negatives in sample classification at cross validation (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e and Additional file 1: Figure \u003cspan refid=\"MOESM5\" class=\"InternalRef\"\u003eS5\u003c/span\u003e). To our knowledge, this represents the first statistically grounded machine learning model that leverages metabolomic data to classify and predict future physical function outcomes in survivors of ARF.\u003c/p\u003e\u003cp\u003eSome metabolites with predictive utility (2-hydroxy-3-methylvalerate, hyocholate, cystine, and β-alanine) showed temporal trends and partial separation between good and poor SPPB groups at day 1 or day 5, but none were considered differential (nominal p\u0026thinsp;\u0026gt;\u0026thinsp;0.05) at these time points (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e and Additional file 1: Figure S7). Independent validation of the predictive performance of the Bayesian logistic regression model is necessary to further strengthen the case for the application of these metabolites to predict PICS of discharged ARF patients.\u003c/p\u003e\u003cp\u003eThe biomarkers identified as predictive of physical function in this study are FDR-independent, as they were selected based on multivariate modeling rather than univariate significance [77,78]. A limitation of this study is that the metabolites identified differential (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) did not retain significance after correction for multiple testing (adjusted p\u0026thinsp;\u0026gt;\u0026thinsp;0.05) (see Additional File 3). This may be partly attributable to the modest sample size. As part of our next step, determining absolute metabolite concentrations through calibration to internal and external standards in a targeted metabolomics approach will help minimize variability in metabolite quantification and with increased sample size and statistical power, thereby strengthening the validation of these findings [79].\u003c/p\u003e\u003cp\u003eThis pilot study was designed to maximize the use of available resources and to guide the identification of metabolites for focus in future, more targeted investigations. We were also limited by the number of patients that had the final 6-month SPPB scores. Future prospective studies may need to utilize different tools for determining physical function, such as activity tracker watches that do not require direct patient interaction after patient discharge. While the predictive modeling and cross-validation results are promising, the absence of an independent validation cohort limits our ability to assess the model\u0026rsquo;s generalizability and stability.\u003c/p\u003e\u003cp\u003eIncreased hospital length of stay among ARF survivors enrolled in this study was significantly associated with poor SPPB scores, indicating worse physical function. This variability in discharge timing may have introduced some heterogeneity into the metabolomic findings, particularly if longer hospitalization reflects differing stages of recovery or persistent metabolic impairment. Importantly, the relatively longer LOS in survivors with poor physical performance aligns with our postulate of a continuum of bioenergetic dysregulation. Non-survivors in our prior studies exhibited severe mitochondrial dysfunction and global metabolic collapse, the survivors with diminished physical function in this cohort may represent an intermediate phenotype, marked by sub-lethal, bioenergetic deficits that impair recovery [29\u0026ndash;34].\u003c/p\u003e\u003cp\u003eAll patients in the study cohort received standard of care which included various medications, that may have some influence on the metabolomic profile. We did not stratify outcomes based on the underlying etiology of ARF (infectious vs. non-infectious), which may have contributed to variability in the observed metabolomic and functional outcomes. This study was retrospective; future prospective studies are necessary to confirm the clinical utility of the predictive model. Targeted metabolomics, particularly for NAD⁺, glycolysis, and mitochondrial redox pathways, may improve sensitivity and specificity in future analyses. Longitudinal monitoring of post-discharge metabolomic profiles would help determine whether these pathways remain significantly altered in patients with poor outcomes and could also assess the impact of interventions such as diet and physical therapy on metabolic recovery. In the next phase of our research, we will conduct prospective validation in independent ARF cohorts across multiple centers with increased sample size. These studies will also incorporate socioeconomic factors and data on access to, and engagement in, physical therapy, thereby providing a more comprehensive understanding of the determinants of physical recovery after ICU discharge.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eMetabolomic profiling of bioenergetic-related metabolites at ICU discharge can distinguish ARF survivors with good versus poor physical performance. We propose that serum metabolomic analysis at discharge offers actionable clinical insight into a patient\u0026rsquo;s metabolic status and recovery trajectory. Quantitative targeted measurement of key bioenergetic and bile acid metabolites may enable risk stratification, inform nutritional or rehabilitative interventions, and personalize post-ICU care. While future prospective studies are required to validate the accuracy and clinical utility of this model, the predictive metabolomic panel identified in this study holds strong potential as a prognostic tool to identify patients at high risk for poor physical recovery and to guide targeted interventions aimed at improving long-term outcomes.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eSPPB\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eshort physical performance battery\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eARF\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eacute respiratory failure\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eUHPLC-MS\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eultra-high performance liquid chromatography\u0026ndash;mass spectrometry\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eICU\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eintensive care unit\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eCOPD\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003echronic obstructive pulmonary disease\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eARDS\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eacute respiratory distress syndrome\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003ePICS\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003epost-intensive care syndrome\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003ePLS-DA\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003epartial least squares discriminant analysis\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eAPACHE\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eAcute Physiology and Chronic Health Evaluation\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eTCA\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003etricarboxylic acid cycle\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eROC\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003ereceiver operating characteristic curve\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003earea under the curve.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAvailability of Data and Materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe dataset supporting the conclusions of this article is included within the article and its additional files. The code used to process and analyze the metabolomics data is publicly available at https://github.com/RNABioUSA/arfqol-metabolomics.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate:\u0026nbsp;\u003c/strong\u003eThis retrospective study involved ARF patients enrolled in the Standardized Rehabilitation for ICU Patients with Acute Respiratory Failure clinical trial at Wake Forest Baptist Medical Center in North Carolina (ClinicalTrials.gov Identifier, NCT00976833; registration date, 2009-09-11). The Wake Forest IRB approval is in accordance with the Declaration of Helsinki. Institutional review board approval was obtained, and informed consent was provided by patients or their legal representatives.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interest:\u0026nbsp;\u003c/strong\u003eWe developed a predictive model using a Bayesian framework that incorporates seven key metabolite biomarkers to estimate future physical function in critically ill patients following hospital discharge. To support future clinical translation, a provisional patent application has been filed to secure intellectual property rights for this biomarker-based prediction method.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot\u0026nbsp;applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAIH, JTR, GTD, RGB data analysis, manuscript preparation. LDP Clinical data collection and data analysis. SG, DCF (D. Clark Files) clinical data analysis and manuscript preparation. EMH, VMP, TS manuscript preparation. DCF, PEM, MNG, and RJL conceptualized the idea, generated the data, guided the development, funding, and revision of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNIH Grants: 1UL1TR001417; 1KL2TR003097; 1 R21 NR019338-01. Center for Lung Biology Frederick P. Whiddon College of Medicine: Gary and Susan Godwin Emerging Scholars Endowed Award; Murray Bander Faculty Development Award. Frederick P. Whiddon College of Medicine Dean\u0026apos;s Predoctoral Fellowship Award. This work was supported in part by an Early Career Investigator Award from the American Thoracic Society (JTR).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eHanley ME, Bone RC. Acute respiratory failure. Pathophysiology, causes, and clinical manifestations. Postgrad Med. 1986;79(1):166\u0026ndash;76. \u003c/li\u003e\n\u003cli\u003eSummers C, Todd RS, Vercruysse GA, Moore FA. Acute Respiratory Failure. In: Perioperative Medicine. Amstardam: Elsevier; 2022. p. 576\u0026ndash;86. \u003c/li\u003e\n\u003cli\u003eHu Q, Hao C, Tang S. From sepsis to acute respiratory distress syndrome (ARDS): Emerging preventive strategies based on molecular and genetic researches. Biosci Rep. 2020;40(5):1\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003ePravda J. Sepsis: Evidence-based pathogenesis and treatment. World J Crit Care Med. 2021;10(4):66\u0026ndash;80. \u003c/li\u003e\n\u003cli\u003eVincent JL, Ak\u0026ccedil;a S, De Mendon\u0026ccedil;a A, Haji-Michael P, Sprung C, Moreno R, et al. The epidemiology of acute respiratory failure in critically III patients. Chest. 2002;121(5):1602\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eGarland A, Dawson N V., Altmann I, Thomas CL, Phillips RS, Tsevat J, et al. Outcomes up to 5 years after severe, acute respiratory failure. Chest. 2004;126(6):1897\u0026ndash;904. \u003c/li\u003e\n\u003cli\u003eDummer J, Stokes T. Improving continuity of care of patients with respiratory disease at hospital discharge. Breathe. 2020;16(3):1\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003ePalakshappa JA, Krall JTW, Belfield LT, Files DC. Long-Term Outcomes in Acute Respiratory Distress Syndrome. Crit Care Clin. 2021 Oct;37(4):895\u0026ndash;911. \u003c/li\u003e\n\u003cli\u003eRolfsen M, Mart MF, Sevin CM, Kieffer H, Krasinski DJ, Ferrante LE, et al. Communication of Post Intensive Care Syndrome: What Providers Reportedly Do and What Patients Remember. Am J Respir Crit Care Med. 2025;211(Abstracts):A1172\u0026ndash;A1172. \u003c/li\u003e\n\u003cli\u003eMeservey AJ, Burton MC, Priest J, Teneback CC, Dixon AE. Risk of Readmission and Mortality Following Hospitalization with Hypercapnic Respiratory Failure. Lung. 2020;198(1):121\u0026ndash;34. \u003c/li\u003e\n\u003cli\u003eAdler D, Peṕin JL, Dupuis-Lozeron E, Espa-Cervena K, Merlet-Violet R, Muller H, et al. Comorbidities and subgroups of patients surviving severe acute hypercapnic respiratory failure in the intensive care unit. Am J Respir Crit Care Med. 2017;196(2):200\u0026ndash;7. \u003c/li\u003e\n\u003cli\u003eBoelens YFN, Melchers M, Van Zanten ARH. Poor physical recovery after critical illness: Incidence, features, risk factors, pathophysiology, and evidence-based therapies. Curr Opin Crit Care. 2022;28(4):409\u0026ndash;16. \u003c/li\u003e\n\u003cli\u003eHerridge MS, Moss M, Hough CL, Hopkins RO, Rice TW, Bienvenu OJ, et al. Recovery and outcomes after the acute respiratory distress syndrome (ARDS) in patients and their family caregivers. Intensive Care Med. 2016;42(5):725\u0026ndash;38. \u003c/li\u003e\n\u003cli\u003eHerridge MS, Tansey CM, Matt\u0026eacute; A, Tomlinson G, Diaz-Granados N, Cooper A, et al. Functional Disability 5 Years after Acute Respiratory Distress Syndrome. N Engl J Med. 2011 Apr 7;364(14):1293\u0026ndash;304. \u003c/li\u003e\n\u003cli\u003eHeesakkers H, Van Der Hoeven JG, Corsten S, Janssen I, Ewalds E, Simons KS, et al. Clinical Outcomes among Patients with 1-Year Survival Following Intensive Care Unit Treatment for COVID-19. Jama. 2022;327(6):559\u0026ndash;65. \u003c/li\u003e\n\u003cli\u003eGerth AMJ, Hatch RA, Young JD, Watkinson PJ. Changes in health-related quality of life after discharge from an intensive care unit: a systematic review. Anaesthesia. 2019;74(1):100\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eOeyen SG, Vandijck DM, Benoit DD, Annemans L, Decruyenaere JM. Quality of life after intensive care: A systematic review of the literature. Crit Care Med. 2010;38(12):2386\u0026ndash;400. \u003c/li\u003e\n\u003cli\u003eWatson RS, Asaro LA, Hutchins L, Bysani GK, Killien EY, Angus DC, et al. Risk factors for functional decline and impaired quality of life after pediatric respiratory failure. Am J Respir Crit Care Med. 2019;200(7):900\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eMayer KP, Welle MM, Evans CG, Greenhill BG, Montgomery-Yates AA, Dupont-Versteegden EE, et al. Muscle Power is Related to Physical Function in Patients Surviving Acute Respiratory Failure: A Prospective Observational Study. Am J Med Sci. 2021;361(3):310\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eWelch SA, Ward RE, Beauchamp MK, Leveille SG, Travison T, Bean JF. The Short Physical Performance Battery (SPPB): A Quick and Useful Tool for Fall Risk Stratification Among Older Primary Care Patients. J Am Med Dir Assoc. 2021;22(8):1646\u0026ndash;51. \u003c/li\u003e\n\u003cli\u003ePavasini R, Guralnik J, Brown JC, di Bari M, Cesari M, Landi F, et al. Short Physical Performance Battery and all-cause mortality: Systematic review and meta-analysis. BMC Med. 2016;14(1):1\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eWestern MJ, Malkowski OS. Associations of the Short Physical Performance Battery (SPPB) with Adverse Health Outcomes in Older Adults: A 14-Year Follow-Up from the English Longitudinal Study of Ageing (ELSA). Int J Environ Res Public Health. 2022;19(23). \u003c/li\u003e\n\u003cli\u003eVoiriot G, Oualha M, Pierre A, Salmon-Gandonni\u0026egrave;re C, Gaudet A, Jouan Y, et al. Chronic critical illness and post-intensive care syndrome: from pathophysiology to clinical challenges. Ann Intensive Care. 2022;12(1). \u003c/li\u003e\n\u003cli\u003eIntensive Care Unit. Intensive Care 2020 and beyond : Co-developing the future. 2020. 6\u0026ndash;15 p. \u003c/li\u003e\n\u003cli\u003eTurnbull AE, Lee EM, Dinglas VD, Beesley S, Bose S, Banner-Goodspeed V, et al. Health Expectations and Quality of Life After Acute Respiratory Failure: A Multicenter Prospective Cohort Study. Chest. 2023;164(1):114\u0026ndash;23. \u003c/li\u003e\n\u003cli\u003eSnowden S, Dahl\u0026eacute;n SE, Wheelock CE. Application of metabolomics approaches to the study of respiratory diseases. Bioanalysis. 2012;4(18):2265\u0026ndash;90. \u003c/li\u003e\n\u003cli\u003eStringer KA, McKay RT, Karnovsky A, Qu\u0026eacute;merais B, Lacy P. Metabolomics and its application to acute lung diseases. Front Immunol. 2016;7(FEB). \u003c/li\u003e\n\u003cli\u003ePatti GJ, Yanes O, Siuzdak G. Metabolomics: the apogee of the omic triology. Nat Rev Mol Cell Biol. 2012;13(4):2504. \u003c/li\u003e\n\u003cli\u003eLangley RJ, Tsalik EL, Van Velkinburgh JC, Glickman SW, Rice BJ, Wang C, et al. Sepsis: An integrated clinico-metabolomic model improves prediction of death in sepsis. Sci Transl Med. 2013;5(195):1\u0026ndash;18. \u003c/li\u003e\n\u003cli\u003eLangley RJ, Migaud ME, Flores L, Thompson JW, Kean EA, Mostellar MM, et al. A metabolomic endotype of bioenergetic dysfunction predicts mortality in critically ill patients with acute respiratory failure. Sci Rep. 2021;11(1):1\u0026ndash;12. \u003c/li\u003e\n\u003cli\u003eRogers AJ, McGeachie M, Baron RM, Gazourian L, Haspel JA, Nakahira K, et al. Metabolomic derangements are associated with mortality in critically ill adult patients. PLoS One. 2014;9(1):1\u0026ndash;7. \u003c/li\u003e\n\u003cli\u003eLangley RJ, Tipper JL, Bruse S, Baron RM, Tsalik EL, Huntley J, et al. Integrative \u0026ldquo;omic\u0026rdquo; analysis of experimental bacteremia identifies a metabolic signature that distinguishes human sepsis from systemic inflammatory response syndromes. Am J Respir Crit Care Med. 2014;190(4):445\u0026ndash;55. \u003c/li\u003e\n\u003cli\u003eLelubre C, Vincent JL. Mechanisms and treatment of organ failure in sepsis. Nat Rev Nephrol. 2018;14(7):417\u0026ndash;27. \u003c/li\u003e\n\u003cli\u003eTsalik EL, Willig LK, Rice BJ, van Velkinburgh JC, Mohney RP, McDunn JE, et al. Renal systems biology of patients with systemic inflammatory response syndrome. Kidney Int. 2015 Oct;88(4):804\u0026ndash;14. \u003c/li\u003e\n\u003cli\u003eSorriento D, Di Vaia E, Iaccarino G. Physical Exercise: A Novel Tool to Protect Mitochondrial Health. Front Physiol. 2021;12(April):1\u0026ndash;14. \u003c/li\u003e\n\u003cli\u003eBrand MD, Orr AL, Perevoshchikova I V., Quinlan CL. The role of mitochondrial function and cellular bioenergetics in ageing and disease. Br J Dermatol. 2013;169(SUPPL.2):1\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eMorris PE, Berry MJ, Files DC, Thompson JC, Hauser J, Flores L, et al. Standardized rehabilitation and hospital length of stay among patients with acute respiratory failure a randomized clinical trial. JAMA - J Am Med Assoc. 2016;315(24):2694\u0026ndash;702. \u003c/li\u003e\n\u003cli\u003eGandotra S, Lovato J, Case D, Bakhru RN, Gibbs K, Berry M, et al. Physical function trajectories in survivors of acute respiratory failure. Ann Am Thorac Soc. 2019;16(4):471\u0026ndash;7. \u003c/li\u003e\n\u003cli\u003eFaquih T, van Smeden M, Luo J, Le Cessie S, Kastenm\u0026uuml;ller G, Krumsiek J, et al. A workflow for missing values imputation of untargeted metabolomics data. Metabolites. 2020;10(12):1\u0026ndash;23. \u003c/li\u003e\n\u003cli\u003eDekermanjian JP, Shaddox E, Nandy D, Ghosh D, Kechris K. Mechanism-aware imputation: a two-step approach in handling missing values in metabolomics. BMC Bioinformatics. 2022;23(1):1\u0026ndash;17. \u003c/li\u003e\n\u003cli\u003eSchiffman C, Petrick L, Perttula K, Yano Y, Carlsson H, Whitehead T, et al. Filtering procedures for untargeted lc-ms metabolomics data. BMC Bioinformatics. 2019;20(1):1\u0026ndash;10. \u003c/li\u003e\n\u003cli\u003eBourgon R, Gentleman R, Huber W. Independent filtering increases detection power for high-throughput experiments. Proc Natl Acad Sci U S A. 2010;107(21):9546\u0026ndash;51. \u003c/li\u003e\n\u003cli\u003eRNABioUSA/arfqol-metabolomics [Internet]. [cited 2025 Jul 24]. Available from: https://github.com/RNABioUSA/arfqol-metabolomics\u003c/li\u003e\n\u003cli\u003eBarupal DK, Fiehn O. Chemical Similarity Enrichment Analysis (ChemRICH) as alternative to biochemical pathway mapping for metabolomic datasets. Sci Rep. 2017;7(1):1\u0026ndash;11. \u003c/li\u003e\n\u003cli\u003ePang Z, Lu Y, Zhou G, Hui F, Xu L, Viau C, et al. MetaboAnalyst 6.0: towards a unified platform for metabolomics data processing, analysis and interpretation. Nucleic Acids Res. 2024;52(W1):W398\u0026ndash;406. \u003c/li\u003e\n\u003cli\u003eEwald JD, Zhou G, Lu Y, Kolic J, Ellis C, Johnson JD, et al. Web-based multi-omics integration using the Analyst software suite. Nature Protocols. Springer US; 2024. \u003c/li\u003e\n\u003cli\u003eLiaw A, Wiener M. The R Journal: Classification and regression by randomForest. R J. 2002;2(3):18\u0026ndash;22. \u003c/li\u003e\n\u003cli\u003eBreiman L. Random Forests. Mach Learn. 2001 Oct;45(1):5\u0026ndash;32. \u003c/li\u003e\n\u003cli\u003eBen G, Jonah G, Imad A, Sam B. rstanarm: {Bayesian} applied regression modeling via {Stan} [Internet]. 2020 [cited 2025 Aug 4]. Available from: https://mc-stan.org/rstanarm\u003c/li\u003e\n\u003cli\u003eVehtari A, Gelman A, Gabry J. Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. Stat Comput. 2017;27(5):1413\u0026ndash;32. \u003c/li\u003e\n\u003cli\u003eMembrez M, Migliavacca E, Christen S, Yaku K, Trieu J, Lee AK, et al. Trigonelline is an NAD+ precursor that improves muscle function during ageing and is reduced in human sarcopenia. Nat Metab. 2024;6(3):433\u0026ndash;47. \u003c/li\u003e\n\u003cli\u003eTraish AM, Abdallah B, Traish AM, Yu G, Traish AM. Androgen deficiency and mitochondrial dysfunction: Implications for fatigue., muscle dysfunction., insulin resistance., diabetes, and cardiovascular disease. Horm Mol Biol Clin Investig. 2011;8(1):431\u0026ndash;44. \u003c/li\u003e\n\u003cli\u003eAhmad I, Newell-Fugate AE. Role of androgens and androgen receptor in control of mitochondrial function. Am J Physiol Cell Physiol. 2022;323(3):C835\u0026ndash;46. \u003c/li\u003e\n\u003cli\u003eHoleček M. Branched-chain amino acids in health and disease: metabolism, alterations in blood plasma, and as supplements. Nutr Metab (Lond). 2018 Dec 3;15(1):33. \u003c/li\u003e\n\u003cli\u003ePinto-Plata V, Casanova C, Divo M, Tesfaigzi Y, Calhoun V, Sui J, et al. Plasma metabolomics and clinical predictors of survival differences in COPD patients. Respir Res. 2019;20(1):1\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eLovell D, Miller D, Capra J, Bradley A. Never mind the metrics -- what about the uncertainty? Visualising confusion matrix metric distributions. 2022; \u003c/li\u003e\n\u003cli\u003eJones GL, Qin Q. Markov Chain Monte Carlo in Practice. Annu Rev Stat Its Appl. 2022;9:557\u0026ndash;78. \u003c/li\u003e\n\u003cli\u003eLauwers C, De Bruyn L, Langouche L. Impact of critical illness on cholesterol and fatty acids: insights into pathophysiology and therapeutic targets. Intensive Care Med Exp . 2023;11(1). \u003c/li\u003e\n\u003cli\u003eOh JH, Chae G, Song JW. Blood lipid profiles as a prognostic biomarker in idiopathic pulmonary fibrosis. Respir Res. 2024;25(1):1\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eZhang D, Zhu Y, Su Y, Yu M, Xu X, Wang C, et al. Taurochenodeoxycholic acid inhibits the proliferation and invasion of gastric cancer and induces its apoptosis. J Food Biochem. 2022;46(3):1\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eBao L, Hao D, Wang X, He X, Mao W, Li P. Transcriptome investigation of anti‐inflammation and immuno‐regulation mechanism of taurochenodeoxycholic acid. BMC Pharmacol Toxicol. 2021;22(1):1\u0026ndash;11. \u003c/li\u003e\n\u003cli\u003eSlijepcevic D, Roscam Abbing RLP, Katafuchi T, Blank A, Donkers JM, van Hoppe S, et al. Hepatic uptake of conjugated bile acids is mediated by both sodium taurocholate cotransporting polypeptide and organic anion transporting polypeptides and modulated by intestinal sensing of plasma bile acid levels in mice. Hepatology. 2017;66(5):1631\u0026ndash;43. \u003c/li\u003e\n\u003cli\u003eSalhab A, Amer J, Lu Y, Safadi R. Sodium + /taurocholate cotransporting polypeptide as target therapy for liver fibrosis. Gut. 2022;71(7):1373\u0026ndash;85. \u003c/li\u003e\n\u003cli\u003eNickler M, Ottiger M, Steuer C, Huber A, Anderson JB, M\u0026uuml;ller B, et al. Systematic review regarding metabolic profiling for improved pathophysiological understanding of disease and outcome prediction in respiratory infections. Respir Res. 2015;16(1). \u003c/li\u003e\n\u003cli\u003eAmasene M, Besga A, Medrano M, Urquiza M, Rodriguez-Larrad A, Tobalina I, et al. Nutritional status and physical performance using handgrip and SPPB tests in hospitalized older adults. Clin Nutr. 2021;40(11):5547\u0026ndash;55. \u003c/li\u003e\n\u003cli\u003eBrydges CR, Fiehn O, Mayberg HS, Schreiber H, Dehkordi SM, Bhattacharyya S, et al. Indoxyl sulfate, a gut microbiome-derived uremic toxin, is associated with psychic anxiety and its functional magnetic resonance imaging-based neurologic signature. Sci Rep. 2021;11(1):1\u0026ndash;14. \u003c/li\u003e\n\u003cli\u003eWeber D, Oefner PJ, Hiergeist A, Koestler J, Gessner A, Weber M, et al. Low urinary indoxyl sulfate levels early after transplantation reflect a disrupted microbiome and are associated with poor outcome. Blood. 2015 Oct 1;126(14):1723\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003evan der Spek A, Stewart ID, K\u0026uuml;hnel B, Pietzner M, Alshehri T, Gau\u0026szlig; F, et al. Circulating metabolites modulated by diet are associated with depression. Mol Psychiatry. 2023;28(9):3874\u0026ndash;87. \u003c/li\u003e\n\u003cli\u003eChakrabarty G, Das S, Dhara P. Impact of nutritional therapy on outcomes in critical care : A review of guidelines and clinical evidence. 2025;7(1):81\u0026ndash;5. \u003c/li\u003e\n\u003cli\u003eSbaih N, Hawthorne K, Lutes J, Cavallazzi R. Nutrition Therapy in Non-intubated Patients with Acute Respiratory Failure. Curr Nutr Rep. 2021;10(4):307\u0026ndash;16. \u003c/li\u003e\n\u003cli\u003eDerave W, \u0026Ouml;zdemir MS, Harris RC, Pottier A, Reyngoudt H, Koppo K, et al. \u0026beta;-Alanine supplementation augments muscle carnosine content and attenuates fatigue during repeated isokinetic contraction bouts in trained sprinters. J Appl Physiol. 2007 Nov;103(5):1736\u0026ndash;43. \u003c/li\u003e\n\u003cli\u003eHarris RC, Tallon MJ, Dunnett M, Boobis L, Coakley J, Kim HJ, et al. The absorption of orally supplied \u0026beta;-alanine and its effect on muscle carnosine synthesis in human vastus lateralis. Amino Acids. 2006;30(3 SPEC. ISS.):279\u0026ndash;89. \u003c/li\u003e\n\u003cli\u003eSchnuck JK, Sunderland KL, Kuennen MR, Vaughan RA. Characterization of the metabolic effect of \u0026beta;-alanine on markers of oxidative metabolism and mitochondrial biogenesis in skeletal muscle. J Exerc Nutr Biochem. 2016;20(2):34\u0026ndash;41. \u003c/li\u003e\n\u003cli\u003eCesak O, Vostalova J, Vidlar A, Bastlova P, Student V. Carnosine and Beta-Alanine Supplementation in Human Medicine: Narrative Review and Critical Assessment. Nutrients. 2023;15(7):1\u0026ndash;21. \u003c/li\u003e\n\u003cli\u003eSaunders B, Elliott-Sale K, Artioli GG, Swinton PA, Dolan E, Roschel H, et al. \u0026beta;-Alanine supplementation to improve exercise capacity and performance: A systematic review and meta-Analysis. Br J Sports Med. 2017;51(8):658\u0026ndash;69. \u003c/li\u003e\n\u003cli\u003eMat\u0026eacute;-Mu\u0026ntilde;oz JL, Lougedo JH, Garnacho-Casta\u0026ntilde;o M V., Veiga-Herreros P, Lozano-Estevan M del C, Garc\u0026iacute;a-Fern\u0026aacute;ndez P, et al. Effects of \u0026beta;-alanine supplementation during a 5-week strength training program: a randomized, controlled study. J Int Soc Sports Nutr. 2018;15(1):1\u0026ndash;12. \u003c/li\u003e\n\u003cli\u003eLo A, Chernoff H, Zheng T, Lo SH. Why significant variables aren\u0026rsquo;t automatically good predictors. Proc Natl Acad Sci U S A. 2015;112(45):13892\u0026ndash;7. \u003c/li\u003e\n\u003cli\u003eGalal A, Talal M, Moustafa A. Applications of machine learning in metabolomics: Disease modeling and classification. Front Genet. 2022;13(November):1\u0026ndash;25. \u003c/li\u003e\n\u003cli\u003eGriffiths WJ, Koal T, Wang Y, Kohl M, Enot DP, Deigner HP. Targeted metabolomics for biomarker discovery. Angew Chemie - Int Ed. 2010;49(32):5426\u0026ndash;45. \u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"respiratory-research","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"rere","sideBox":"Learn more about [Respiratory Research](http://respiratory-research.biomedcentral.com/)","snPcode":"12931","submissionUrl":"https://submission.nature.com/new-submission/12931/3","title":"Respiratory Research","twitterHandle":"@RespiratoryBMC","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Acute respiratory failure, post intensive care syndrome, physical function, metabolite feature selection for patient classification, logistic regression.","lastPublishedDoi":"10.21203/rs.3.rs-7394034/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7394034/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIntroduction: Acute respiratory failure (ARF) often leads to post-intensive care syndrome, including persistent physical impairments after ICU discharge. Emerging evidence suggests that mitochondrial bioenergetic dysfunction, detectable through metabolomic profiling, may contribute to poor recovery.\u003c/p\u003e\u003cp\u003eMethods: We performed a retrospective study comprising of untargeted metabolomic profiling using ultrahigh performance liquid chromatography\u0026ndash;mass spectrometry (UHPLC-MS) on serial serum samples from 70 ARF patients taken at ICU admission, during hospitalization and at discharge. Physical function was assessed post-discharge using the Short Physical Performance Battery (SPPB). Correlation and logistic regression analyses were performed to identify metabolomic predictors of six-month physical function outcomes.\u003c/p\u003e\u003cp\u003eResults: Patients with poor SPPB scores exhibited dysregulation in bioenergetic metabolite levels, as well as fatty acid oxidation, glycerophospholipid metabolism, bile acid biosynthesis and amino acid metabolism. These metabolic changes were not explained by initial disease severity (APACHE III scores) or comorbidities. In contrast, several metabolites measured at discharge were predictive of SPPB scores with an AUROC of 0.88 after cross validation.\u003c/p\u003e\u003cp\u003eConclusion: Our findings highlight persistent metabolic dysfunction at discharge, particularly in pathways related to bioenergetics. To our knowledge, this is the first study to employ a metabolite-based machine learning model to predict ARF survivors physical function outcomes using serum metabolites measured at discharge. Further insights on dysregulated pathways suggest that nutritional interventions targeting these metabolic pathways, such as supplementation with β-alanine, could potentially improve post-ICU recovery outcomes.\u003c/p\u003e","manuscriptTitle":"Metabolomic Biomarkers Predict Long-term Physical Function in Survivors of Acute Respiratory Failure","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-02 14:55:37","doi":"10.21203/rs.3.rs-7394034/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Accepted","date":"2025-11-06T03:08:58+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-03T09:58:06+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"59401127047575881844754348643214489860","date":"2025-10-31T14:41:44+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-31T13:40:49+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"205661145315837897381485482668629476115","date":"2025-10-31T13:31:10+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-30T15:44:24+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-30T13:15:34+00:00","index":"","fulltext":""},{"type":"submitted","content":"Respiratory Research","date":"2025-10-27T16:26:52+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"respiratory-research","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"rere","sideBox":"Learn more about [Respiratory Research](http://respiratory-research.biomedcentral.com/)","snPcode":"12931","submissionUrl":"https://submission.nature.com/new-submission/12931/3","title":"Respiratory Research","twitterHandle":"@RespiratoryBMC","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"7420a44a-294f-47ee-9e0b-ecc62d485d7f","owner":[],"postedDate":"November 2nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-11-24T16:01:21+00:00","versionOfRecord":{"articleIdentity":"rs-7394034","link":"https://doi.org/10.1186/s12931-025-03423-2","journal":{"identity":"respiratory-research","isVorOnly":false,"title":"Respiratory Research"},"publishedOn":"2025-11-21 15:57:08","publishedOnDateReadable":"November 21st, 2025"},"versionCreatedAt":"2025-11-02 14:55:37","video":"","vorDoi":"10.1186/s12931-025-03423-2","vorDoiUrl":"https://doi.org/10.1186/s12931-025-03423-2","workflowStages":[]},"version":"v1","identity":"rs-7394034","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7394034","identity":"rs-7394034","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}