Critically ill patients admitted to the intensive care unit (ICU) from the emergency department (ED) represent a population at high risk of adverse outcomes, with mortality rates significantly influenced by the timeliness and appropriateness of initial management (1, 2). Among these, patients requiring surgical ICU admission face unique challenges due to the interplay of acute surgical stress, underlying comorbidities, and the metabolic consequences of critical illness (3, 4). The ED is a major source of ICU admissions, with surgical ICU mortality rates ranging from 1.3 to 7.56% (5–7). Traditional prognostic tools, such as the Acute Physiology and Chronic Health Evaluation (APACHE) IV and the Sequential Organ Failure Assessment (SOFA) score (8, 9), though valuable, often incorporate parameters that are not readily available during the initial ED assessment, limiting their utility for refining prognosis and guiding subsequent fluid-nutrition management.

Body mass index (BMI), the cornerstone of current obesity classification systems, is widely used to reflect nutritional and metabolic status because of its simplicity and broad availability (10). The relationship between BMI and critical-care outcomes has been extensively studied, often manifesting as an “obesity paradox,” in which overweight or mildly obese patients experience lower mortality than underweight or normal-weight individuals (11–13). However, the vast majority of this evidence is based on a single, static BMI measurement obtained at hospital admission. This static snapshot fails to capture the dynamic physiological changes that occur during hospitalization, such as fluid shifts, catabolic states, and the effects of nutritional support, all of which can significantly alter body composition and potentially influence prognosis (14).

This is particularly relevant in surgical ICU populations. The metabolic response to surgical trauma and critical illness can lead to rapid changes in weight, reflecting either fluid retention or lean body mass catabolism (15, 16). In the acute phase of critical illness, particularly in surgical patients, changes in body weight and BMI are predominantly influenced by fluid shifts and balance, rather than changes in fat or lean mass. Thus, to some extent, an increase in BMI can be regarded as a surrogate for fluid overload—a well-established risk factor for organ dysfunction and increased mortality in the ICU (17). Conversely, a decrease in BMI might reflect inadequate caloric intake or severe catabolism, also potentially leading to poorer outcomes (18). Despite the biological plausibility of these mechanisms, the prognostic value of dynamic, in-hospital BMI change has not been thoroughly investigated in surgical patients admitted to the ICU from the emergency setting.

To address this significant knowledge gap, we conducted a large, retrospective cohort study utilizing the eICU Collaborative Research Database (19). Our primary objective was to investigate the association between the change in BMI during hospitalization and 28-day all-cause mortality in a cohort of surgical ICU patients admitted from the ED. We further aimed to explore the nature of this association (linear vs. non-linear) and to evaluate whether dynamic BMI change offers superior prognostic value compared to static admission or discharge BMI measurements.

This large multicenter retrospective cohort study demonstrates that in-hospital BMI change is a significant and independent predictor of 28-day ICU mortality among surgical ICU patients admitted via the ED. We identified a clear non-linear, U-shaped relationship between BMI change and mortality risk, with the lowest mortality observed at a BMI reduction of approximately 1.75 kg/m². Moreover, comparative analyses of body-mass metrics revealed that dynamic BMI change during hospitalization offers superior predictive performance for 28-day mortality compared with static weight or BMI measurements at either admission or discharge, underscoring its potential utility in clinical risk assessment for this vulnerable population.

In the present study, the observed 28-day mortality was 4.7%, closely resembling the 5.42% reported by Falcão et al. (2019) in surgical ICU patients (6), thereby reinforcing the external validity of our findings. To our knowledge, mortality in surgical ICUs is generally lower than in medical or mixed ICUs, primarily because of differences in case-mix: the latter cohorts are dominated by older patients with multiple comorbidities (23, 24). When stratified by quartiles of in-hospital BMI change (Q1 to Q4), a clear gradient in mortality risk was observed, rising from 3.77% in Q1 (greatest BMI decrease) to 7.27% in Q4 (greatest BMI increase), as detailed in Table 1. This mortality rate is comparatively lower than that reported in several previous studies focusing on mixed or medical ICU populations. The relatively lower mortality in our study may be attributable to the specific inclusion of surgical patients admitted through the ED, who might benefit from timely surgical interventions and targeted perioperative management. Furthermore, the exclusion of patients with extreme BMI changes (exceeding mean ± 3 SD) or short ICU stays (<24 h) likely contributed to a more stable cohort with a different risk profile compared to general ICU populations.

The relationship between baseline body composition, as crudely estimated by admission BMI, and mortality in critically ill patients remains complex and at times paradoxical. While some studies, particularly in medical ICU populations, have reported the existence of an “obesity paradox” where overweight or obese patients exhibit lower mortality rates compared to underweight or normal-weight individuals (25, 26), our findings in a surgical ICU cohort admitted via the ED did not identify admission BMI or discharge BMI as significant predictors of 28-day mortality in univariate analysis. Moreover, we compared the prognostic value of various BMI-related indices for 28-day mortality in surgical ICU patients and found that in-hospital BMI change exhibited the highest discriminatory performance, as detailed in Supplementary Table S4. This suggests that a single, static measure of body composition at a single time point may be insufficient to capture the dynamic metabolic and fluid shifts that truly impact prognosis in acute surgical settings (27). In contrast, our study establishes that the change in BMI during hospitalization is a robust and independent predictor of mortality. Even after comprehensive adjustment, the positive association persisted. Patients in the highest quartile of BMI increase (Q4) faced a 52% greater risk of mortality compared to those in the lowest quartile (Q1). This result is consistent with a growing body of literature linking positive fluid balance and weight gain during ICU stay to impaired organ function and increased mortality (28, 29), likely reflecting pathological fluid overload rather than anabolic gain. The discrepancy between our findings and those reporting the obesity paradox may be attributed to key differences in study population—specifically, the inclusion of emergency surgical patients experiencing acute physiological stress—and, most importantly, the focus on dynamic change rather than a static baseline measurement, which more accurately captures the deleterious impact of fluid accumulation and metabolic dysregulation on survival (30).

To further elucidate the true nature of the association between in-hospital BMI change and 28-day mortality, we employed restricted cubic spline (RCS) analysis. The results revealed a significant non-linear U-shaped relationship (P for non-linearity < 0.001). This finding contrasts sharply with the linear association reported by Zhang et al. in their study using the same database, which suggested that any increase in BMI was associated with increased mortality risk (31). Threshold analysis precisely identified an inflection point at −1.75 kg/m², indicating the lowest mortality risk around this moderate degree of weight reduction. Mortality risk increased sharply when BMI rose above this threshold, while showing a gradual increase when BMI decrease exceeded −1.75 kg/m² (Table 3). This U-shaped pattern aligns well with findings from critical illness fluid balance studies. Shen Y et al. (32) demonstrated that achieving negative fluid balance (or mild weight reduction) through deresuscitation strategies in the later phases of critical illness was associated with improved survival, potentially through mitigating organ dysfunction related to fluid overload. Concurrently, our observation that significant weight gain (right arm of the U-curve) correlates with higher mortality is supported by extensive literature linking positive fluid balance with adverse outcomes. The optimal BMI reduction identified in our study (approximately 1.75 kg/m²) is more pronounced than values reported in some medical ICU studies. This discrepancy primarily stems from two factors. First, the study populations differ: our cohort specifically comprised surgical patients admitted via the ED, who experience more pronounced third-space fluid shifts and greater inflammatory insults, making them particularly susceptible to fluid overload. Second, the methodological approaches differ: unlike the linear model used by Zhang et al. (31), our flexible modeling strategy (restricted cubic splines) more accurately captured the underlying non-linear relationship. Therefore, for this specific surgical population, a moderate reduction in BMI likely reflects successful fluid mobilization and is associated with optimal clinical outcomes.

The most plausible explanation for the strong association between BMI increase and mortality in our cohort is fluid overload. In critically ill surgical patients, significant weight gain over a short period is far more likely to represent pathological fluid accumulation rather than anabolic growth. This interpretation is corroborated by consistent laboratory patterns across BMI-gain quartiles: the stepwise decrease in RBC and Hb, accompanied by a concurrent rise in WBC (Table 1), aligns with haemodilution and inflammatory capillary leak—rather than blood loss or malnutrition. Patients with the greatest BMI gain (Q4) present higher illness-severity scores (APACHE IV, SOFA) and greater requirement for organ-support therapies (mechanical ventilation, dialysis), mirroring exactly the subgroup in which positive fluid balance is both most frequent and most harmful. Although the eICU-CRD contains no intake/output records, the alignment of physiologic rationale with consistent clinical and laboratory findings reinforces the inference that BMI change captures clinically meaningful fluid accumulation.

Furthermore, clinical correlates strongly support this mechanism. Patients in the highest BMI-change quartile (Q4) exhibited higher illness severity scores (APACHE IV, SOFA) and increased need for organ support (ventilation, dialysis)—clinical contexts in which fluid overload is frequently observed and deleteriously influences outcomes. Thus, our results align with a substantial body of literature linking positive fluid balance to organ dysfunction and increased mortality. Conversely, the elevated mortality observed at the left end of the U-shaped curve, where extreme BMI reduction occurs, may reflect severe catabolism and loss of lean body mass. This pattern underscores the dual interpretation of in-hospital BMI change: it serves as a marker of fluid status in the positive range and potentially of nutritional depletion in the negative range. COPD was the only subgroup showing a significant interaction (p = 0.037). Its systemic inflammatory state and increased capillary leak accelerate fluid accumulation, while pulmonary hyper-inflation and reduced right-ventricular preload render these patients prone to right-heart failure and increased work of breathing even after mild fluid loading (33, 34). Consequently, for any given rise in BMI, mortality climbs more steeply in COPD patients.

This study possesses several notable methodological strengths and offers significant clinical implications. First, it is a large multinational cohort study utilizing data from multiple centers, which enhances the generalizability and statistical power of our findings. Second, we employed multiple imputation to handle missing data, reducing potential bias and preserving the integrity of the dataset. Third, the association remained robust after extensive adjustment for a comprehensive set of clinically relevant covariates, including demographics, and severity scores, strengthening the validity of our conclusions. Fourth, we conducted detailed subgroup analyses and tested for interactions, which demonstrated the consistency of the association across various patient strata and increased the granularity of our results. Fifth, the identification of a specific, clinically actionable threshold for BMI change provides a potential target for goal-directed fluid and nutritional management in this population. These findings suggest that monitoring in-hospital BMI dynamics could serve as a simple, practical tool for improving risk stratification and guiding personalized care strategies for emergently admitted surgical ICU patients.

This study has several limitations. First, the primary exposure, in-hospital BMI change, is interpreted as a surrogate for fluid balance, a limitation inherent to databases lacking direct fluid intake/output documentation. Although the clinical and laboratory correlates strongly support this interpretation, we cannot definitively partition the observed weight change into its fluid, catabolic, and anabolic components. The database does not capture nutrition route, surgical complexity, or specific fluid protocols. We partially adjusted for these factors by including glucocorticoid use, dialysis, ventilator days, and APACHE-IV (which incorporates major surgery), yet unrecorded peri-operative confounders may persist. Relatedly, the timing of the “discharge weight” measurement was not standardized, representing the last available value prior to an endpoint. Second, BMI change serves only as a surrogate for fluid balance and nutritional status and cannot distinguish whether weight variations stem from fluid retention, muscle catabolism, or alterations in fat mass. We lack body composition data (e.g., bioelectrical impedance analysis) to clarify the underlying mechanisms. Furthermore, as U. S. ICUs do not routinely perform weighing at the time of death and the eICU-CRD lacks standardized pre-discharge weighing timepoints, missing weight records may introduce survivor bias. Third, the data originate from the U. S.-based eICU database with a predominantly Caucasian population, which may limit the generalizability of our findings to other ethnic groups or healthcare systems (e.g., Asian populations). Fourth, the exclusion of patients with BMI changes exceeding 15% or ICU stays shorter than 24 h may introduce selection bias and restrict the applicability of our results to extremely critical or short-stay populations. Fifth, although ROC analysis indicated that dynamic BMI change outperformed static measures, its AUC remained modest (57.92%), suggesting that integrating additional clinical variables is necessary to enhance prognostic accuracy. It is important to emphasize that BMI does not capture acute changes in lean body mass or fluid balance; therefore, any inference regarding “nutritional status” based solely on BMI should be considered approximate and requires validation with clinical and biochemical indicators. Future studies should integrate hourly ICU fluid balance records with validated malnutrition screening tools (e.g., NRS-2002) and bedside body composition markers (e.g., mid-upper arm circumference, handgrip strength, bioelectrical impedance) to differentiate fluid-induced weight gain from true muscle loss. As this study utilized discharge BMI to calculate in-hospital change, this metric can only be generated retrospectively and cannot serve as an early predictor at ICU admission. Whether serial bedside weight measurements converted into dynamic BMI trajectories can provide real-time risk alerts warrants evaluation in prospective studies equipped with automated weighing systems and electronic alert mechanisms. Importantly, the spline-determined nadir (≈ − 1.75 kg m⁻²) is a statistical optimum, not a biologically immutable threshold. Consequently, the figure should be viewed as a risk-range indicator rather than a precise therapeutic goal. Randomized trials are required to verify whether deliberate modulation of fluid balance to this zone translates into mortality benefit.

This multicenter retrospective cohort study demonstrates a significant non-linear U-shaped relationship between in-hospital BMI change and 28-day mortality, with the lowest risk observed at a BMI reduction of approximately 1.75 kg/m². Dynamic BMI change during hospitalization offers superior prognostic value compared with static BMI or weight measured at admission or discharge. Serial BMI monitoring can refine risk stratification and guide individualized fluid and nutritional management, thereby holding important clinical value.