This is a single-center retrospective observational study on patients with sepsis admitted to the ICU of a university hospital from January 2014 to December 2021. Sepsis was diagnosed according to the sepsis-3 criteria [22]; in brief, patients with documented or suspected infection and an acute change in total Sequential Organ Failure Assessment (SOFA) score of ≥ 2 points were considered to have sepsis. Patients were eligible if they were 18 years of age or older, and at least 5 days of ICU stay, had undergone an abdominal CT examination within 24 h of admission, and had at least two serial CT datasets during hospitalization, including the abdomen. We excluded patients if (1) they had creatinine > 4 mg/dl on ICU admission, (2) received renal replacement therapy (RRT), (3) There was gastrointestinal bleeding within the three days before admission, (4) failed to match to complete data records, (5) transferred from another ICU, and (6) the interval between two serial CT datasets shorter than 5 days. The study was approved by the ethics committees of Zhongda Hospital, Affiliated to Southeast University (Number 2021ZDSYLL225-P01) and performed in accordance with STROBE [23] guideline for observational research.

For all patients, the following demographic and clinical data were collected from electronic health records: sex, age, height and weight, sequential organ failure assessment (SOFA) score, acute physiology and chronic health evaluation (APACHE) II score, chronic comorbidities, admission diagnosis, and infection site. We recorded the creatine kinase, myoglobin, hemoglobin, albumin (ALB), prognostic nutrition index (PNI), blood glucose, lymphocyte count, procalcitonin (PCT), and lactate on day 1 after ICU admission. Blood urea nitrogen (BUN), and serum creatinine were extracted on day 1 and day 3. In addition, we collected the mode of nutritional support, daily non-protein caloric intake, neuromuscular blockers use, Richmond agitation-sedation scale (RASS) score, and insulin dose. PNI was calculated as 10 × serum ALB level (g/L) + 0.005 × total lymphocyte count (/mm3).

CT images of the third lumbar vertebra (L3) were evaluated. For all patients included, skeletal muscle area at L3 slices (L3SMA) includes the psoas, erector spine, quadratus lumborum, transverse abdominous, external and internal obliques, and rectus abdominous. Images were analyzed by a trained specialist (J.J) with Image J software [6] version 1.48 (https://imagej.nih.gov/ij/index.html; NIH, Bethesda, MD, USA), which used predetermined thresholds (−29 to + 150 Hounsfield units) to demarcate skeletal muscle tissues [6]. Tissue boundaries were manually corrected if needed. L3SMA was calculated automatically by summing the skeletal muscle-tissue pixels and multiplying them by the surface area of each pixel. Figure 1 shows the before and after changes in L3SMA (red) on transverse CT images in the same patient. L3SMA was adjusted for the square of the height (SMA/height2), which was referred to as the skeletal muscle index (L3SMI). L3SMI = L3SMA/ height 2. L3SMA was assessed on ICU admission and at least one more in-ICU follow-up CT. If the patient had ≥ 2 upper abdominal CT scans during the ICU stay, only the first and second CT scan results were evaluated. In Fazzini's (2023) meta-analysis [24], four studies used CT methods to measure the L3SMA, with data provided by three of these studies. Lambell [25] reported a 3.1%/day decrease in L3SMA, Haines [19] found a decrease of 2.1%/day, and Jung [26] observed a 0.8%/day decrease in L3SMA. Based on these findings, along with data from other muscle ultrasound studies, a threshold of 2%/day was set. We define rapid muscle loss as a change in L3SMA (△L3SMA) > 2% per day, calculated from the difference between admission and follow-up CT L3 muscle areas, as follows:

$$\Delta\text3\text=\frac3\text-\text3\text}3\text}$$

Typical transverse CT images at L3 of the same patient. a ICU admission; b, the tenth day after ICU admission. The total skeletal muscle area (red)

The primary exposures were UCR on day 1 (UCR_D1), which was calculated as [urea nitrogen*1000/ creatinine], UCR on day 3 (UCR_D3), and the difference in UCR between day 3 and day 1 (ΔUCR). The primary outcome was the incidence of rapid muscle loss during ICU hospitalization. The secondary outcomes included ICU and hospital length of stay, ICU mortality, 28-day mortality, hospital mortality, and ventilation-free days (VFDs) on day 28.

Values are presented as the mean (standard deviation) or median [interquartile range (IQR)] for continuous variables as appropriate and as the total number (percentage) for categorical variables. Comparisons between rapid muscle loss and no rapid muscle loss patients were made using the X2 test or Fisher’s exact test for categorical variables and Student’s t-test or Mann–Whitney U test for continuous variables as appropriate. The Shapiro–Wilk test was used to assess the normality of continuous variables.

A Spearman rank correlation test was conducted to assess the relationship between ΔUCR and ΔL3SMA.We first employed three multivariable logistic regression models to explore the association between UCR or ΔUCR and rapid muscle loss. Variables based on previous studies and clinical correlations including age, gender, BMI, SOFA score, Lactate, PCT, L3SMA, and PNI were entered into the model. We calculated the area under curve (AUC) of the receiver operating characteristic (ROC) curve to quantify the performance of UCR or ΔUCR in predicting rapid muscle loss. The optimal cut-off value of UCR or ΔUCR was determined by the Youden Index (sensitivity + specificity −1). We also calculated the AUROCs after adjusting for the above confounders. The AUCs were compared by using the bootstrap test for each two ROC curves.

Several subgroup analyses were performed according to sex, age (≥ 60, < 60), Diabetes mellitus, BMI (≥ 24, < 24), SOFA score (≥ 8, < 8), PNI (≥ 34, < 34), L3SMI (≥ 40, < 40), use of mechanical ventilation, insulin, nutrition supports (whether the patient fasted within 3 days of admission), neuromuscular blocker and vasopressor.

We used Stata software version 15.0 (Stata Corp) and R software (IBM, Armonk, NY version 4.0.3) for all analyses. 2-tailed P < 0 0.05 was considered to be statistically significant.