Association between body mass index and mortality in intensive care unit patients with sepsis: A retrospective cohort study Authors: Yunya Zhu, Wenyuan Zhang, Yuting Zhong, Luwen Liang, Ying Li Published: 2026 ABSTRACT AbstractThe link between body mass index (BMI) and outcomes in intensive care unit (ICU) patients with sepsis is inconsistently supported by the available data. This study aimed to examine the association between BMI and mortality among ICU patients with sepsis. On the basis of their BMI values upon admission to the ICU, sepsis patients from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database were categorized as low weight, normal weight, overweight, or obesity group. The primary outcome measure was in-hospital mortality and the secondary outcome was 90-day mortality. Multivariable logistic regression was used to examine the association between BMI categories and in-hospital mortality. Kaplan-Meier methods with log-rank testing were used to compare 90-day survival across groups, and Cox proportional hazards models were applied to estimate the association between BMI and 90-day mortality. Subgroup analyses were also performed. In an external cohort, inverse probability of treatment weighting (IPTW) was applied as a supplementary analysis, and propensity score matching (PSM) was performed as a sensitivity analysis to assess the robustness of the observed associations. A total of 13,864 patients with sepsis were included (low weight: 339; normal weight: 3,582; overweight: 4,564; obesity: 5,379). Overall in-hospital mortality was 15.4%. In multivariable logistic regression analyses, compared with the normal weight group, the adjusted odds ratios (ORs) for in-hospital mortality were 1.68 (95% confidence interval [CI]: 1.29–2.20) in the low weight group, 0.77 (95% CI: 0.67–0.88) in the overweight group, and 0.74 (95% CI: 0.65–0.85) in the obesity group. The overall 90-day mortality was 22.9%. In multivariable Cox proportional hazards models, compared with the normal weight group, the adjusted hazard ratios (HRs) for 90-day mortality were 1.62 (95% CI: 1.34–1.94) in the low weight group, 0.78 (95% CI: 0.71–0.85) in the overweight group, and 0.74 (95% CI: 0.68–0.81) in the obesity group. Subgroup analyses by age and sex showed consistent associations. In the external cohort, a BMI ≥ 32 kg/m² was also associated with lower in-hospital and 90-day mortality after adjustment. In conclusion, BMI was associated with mortality among ICU patients with sepsis. In this retrospective cohort, higher BMI was associated with lower in-hospital and 90-day mortality compared with low or normal BMI after adjustment. These findings are observational and should not be interpreted as evidence of a protective effect or causality.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-54293-7.Keywords: Body mass index, Sepsis, Intensive care unit, mortality, Subgroup analysis Subject terms: Diseases, Health care, Medical research, Risk factors Introduction Sepsis is a fatal syndrome of physiologic, pathologic, and metabolic abnormalities caused by infection, which is one of the most significant global public health issues1. It is estimated that at least 3 million patients around the world suffer from sepsis each year, with fatality rates as high as 30–50%, causing significant harm to not just the victims’ families but also to society as a whole2–4. Therefore, examination of potential risk factors for sepsis and sepsis-related mortality may facilitate sepsis prevention and screening of at-risk individuals for early diagnosis and treatment, therefore reducing the burden of sepsis. Moreover, the proportion of obese patients hospitalized to critical care units (ICUs) is on the rise and creates difficult challenges5,6. Most health organizations, including the WHO, utilize the body mass index (BMI), which is computed by dividing weight by the square of height, as a screening tool for identifying obesity7,8. Underweight, overweight, and obesity are all related with higher mortality in the general population compared to those of normal weight, according to available research9,10. However, some studies have reported an inverse association between overweight or obesity and mortality in certain acute and severe illnesses, including sepsis11–14. In addition, the relationship between BMI and long-term outcomes is debatable, with potential confounding variables including preexisting comorbidities and additional life support needs15. As a result, we decided to conduct a large retrospective cohort study using the Medical Information Mart for Intensive Care IV (MIMIC-IV) database to investigate the association between different BMI categories and in-hospital as well as 90-day mortality among patients with sepsis. Our goal was to assist medical professionals in accurately assessing the condition in a timely manner and provide a foundation for subsequent prognostic measures. Methods Study design Using the MIMIC-IV, a large database in the United States, we conducted a retrospective cohort analysis16. The MIMIC- IV database at Beth Israel Deaconess Medical Center contains extensive and high-quality information on patients admitted to the intensive care units (ICUs). One of the authors, ZWY, was able to successfully finish the database training and obtain the certificate (certification number 41767848). Individual patient consent was waived because the experiment did not alter clinical care and all protected health information was anonymized. Selection of participants Patients in the MIMIC-IV database who met the sepsis criteria were considered eligible for inclusion. Sepsis was defined according to the Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3)17. Briefly, sepsis was identified in patients with proven or suspected infection and an acute increase in the Sequential Organ Failure Assessment (SOFA) score of ≥ 2 points. In MIMIC-IV, infection was identified using the International Classification of Diseases, Ninth Revision (ICD-9) codes. Only data from the first hospital/ICU admission were included for patients with multiple admissions. Patients without height or weight records and those with an ICU stay of less than 24 h were excluded. In addition, only adult patients aged 18 years or older were included, and pediatric patients (< 18 years) were excluded. Patients were categorized according to BMI at ICU admission as follows: low weight (< 18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25–29.9 kg/m²), and obesity (≥ 30 kg/m²)18. The patient selection process is shown in Fig. 1. Screening of admissions for inclusion. External cohort analysis We reviewed data from 526 septic patients admitted to the ICU of the First Affiliated Hospital of Gannan Medical University between January 2010 and December 2024. Consistent with the inclusion criteria of the primary cohort, only adult patients aged 18 years or older were enrolled, and pediatric patients were excluded. Among these, 28 patients were excluded because of missing clinical data, ICU stay of less than 24 h, multiple ICU admissions, loss to follow-up, or other confounding conditions considered unsuitable for inclusion in this study. Ethical approval was obtained from the Institutional Review Board of the First Affiliated Hospital of Gannan Medical University. Given the retrospective nature of the study, the requirement for informed consent was waived by the Ethics Committee. This waiver was consistent with the Declaration of Helsinki and national guidelines for secondary research involving anonymized health data. Variable extraction Using structured query language (SQL), baseline characteristics within the first 24 h after ICU admission were obtained, including age, gender, marital status, insurance, BMI, and admission type. Severity at admission as measured by SOFA score the Simplified Acute Physiology Score II (SAPS II), and Glasgow Coma Scale (GCS). The use of mechanical ventilation (MV), administration of vasoactive agent, extracorporeal membrane oxygenation (ECMO), and application of continuous renal replacement therapy (CRRT) were also recorded. Comorbidities including hypertension, diabetes, myocardial infarction (MI), congestive heart failure (CHF), cerebrovascular disease, chronic obstructive pulmonary disease (COPD), liver disease, renal disease, malignancy, and acquired immune deficiency syndrome (AIDS) were also collected for analysis based on the recorded ICD-9 codes in the MIMIC-IV database. Vital signs included the heart rate (HR), respiratory rate (RR), mean arterial pressure (MAP), and mean temperature. Laboratory variables including white blood cell (WBC) count, haematocrit (HCT), haemoglobin (HGB), platelet (PLT) count, prothrombin time (PT), partial thrombin time (PTT), international normalized ratio (INR), sodium, potassium, calcium, bicarbonate, anion gap (AG), blood glucose (Glu), creatinine (Cr), blood urea nitrogen (BUN), and urine output were measured during the first 24 h in the ICU. If a variable was recorded more than once in the first 24 h, we used the value related to the greatest severity of illness. Outcomes The primary outcome measure was in-hospital mortality and the secondary outcome was 90-day mortality. Statistical analysis No missing data were observed for categorical variables. For continuous variables, the proportion of missing data was generally low (< 10% for variables included in the main analysis). Therefore, missing values were imputed using mean substitution (Table S1). This approach was chosen given the relatively low level of missingness and to preserve sample size and statistical power in this large cohort. Variables with a high proportion of missing data (e.g., lactate and other laboratory indicators with substantial missingness) were not included in the primary analysis to avoid introducing additional bias from excessive imputation. The distribution of continuous variables was assessed using the Anderson-Darling test. Continuous variables are presented as medians with interquartile ranges (IQRs), and categorical variables as counts and percentages. Group comparisons were performed using the χ² test for categorical variables and the Mann-Whitney U test for continuous variables, as appropriate. Restricted cubic spline (RCS) analyses were initially used to explore the potential nonlinear association between BMI and mortality. Multivariable logistic regression models were used to examine the association between BMI categories and in-hospital mortality. Kaplan-Meier methods and the log-rank test were used to compare 90-day survival across BMI groups, and Cox proportional hazards regression models were applied to estimate the association between BMI and 90-day mortality. Based on the RCS results, BMI in the external cohort was further dichotomized using the identified threshold value. In the external cohort, inverse probability of treatment weighting (IPTW) was performed as a supplementary analysis to examine the association between a high BMI threshold and clinical outcomes while balancing measured baseline covariates. Stabilized weights were calculated from the propensity score, and extreme weights were trimmed at the 1st and 99th percentiles. Covariates included in the propensity score model comprised all baseline variables. Covariate balance before and after weighting was assessed using standardized mean differences (SMDs), with an SMD < 0.1 indicating acceptable balance. In addition, propensity score matching (PSM) was performed as a sensitivity analysis to test the robustness of the IPTW findings. Patients in the high-BMI threshold group and non-high-BMI threshold group were matched on the basis of the estimated propensity score using a nearest-neighbor matching algorithm with a caliper of 0.2 standard deviations of the logit of the propensity score and a matching ratio of 1:2. Covariate balance after matching was also evaluated using SMDs. Outcomes in the matched cohort were analyzed using conditional logistic regression for in-hospital mortality and Cox proportional hazards models or Kaplan-Meier analysis for 90-day mortality, as appropriate. All statistical analyses were performed using R software (version 4.1.1), and a two-sided P < 0.05 was considered statistically significant. Results Participant characteristics The MIMIC-IV identified 34,899 patients with sepsis, of whom 13,864 matched the inclusion criteria (Fig. 1). Of the patients included in the study, 339 patients had a BMI < 18.5 kg/m² (low weight group), 3582 patients had a BMI between 18.5 and 25 kg/m² (normal weight group), 4564 patients had a BMI between 25 and 30 kg/m² (overweight group), and 5379 patients had a BMI ≥ 30 kg/m² (obesity group). The median age of the study patients was 67.37 (56.46, 77.48) years old, and 8,409 (60.7%) were males. Table 1 lists the demographic and clinical features of all participants, and the BMI of each group in the research varied significantly (P < 0.001). Table 1Baseline characteristics of subjects in the different BMI levels group.VariablesLow weight group (N = 339)Normal weight group (N = 3582)Overweight group (N = 4564)Obesity group (N = 5379) P Characteristics Age (Years)70.66 (59.16, 83.06)69.42 (56.44, 81.00)68.64 (57.28, 78.47)65.30 (55.74, 73.97)< 0.001Male (%)144 (42.5%)2048 (57.2%)3037 (66.5%)3180 (59.1%)< 0.001Married (%)98 (28.9%)1560 (43.6%)2313 (50.7%)2551 (47.4%)< 0.001Insurance (%)208 (61.4%)1926 (53.8%)2318 (50.8%)2638 (49.0%)< 0.001Admission type (%)< 0.001Emergency223 (65.8%)2290 (63.9%)2812 (61.6%)3169 (58.9%)Selective47 (13.9%)514 (14.3%)674 (14.8%)783 (14.6%)Urgent69 (20.4%)778 (21.7%)1078 (23.6%)1427 (26.5%) Severity of illness SOFA score3.00 (2.00, 4.00)3.00 (2.00, 4.00)3.00 (2.00, 5.00)3.00 (2.00, 5.00)< 0.001SAPS II score43.00 (33.00, 54.00)39.00 (31.00, 49.00)38.00 (30.00, 49.00)38.00 (30.00, 48.00)< 0.001GCS12.00 (8.00, 14.00)13.00 (8.00, 14.00)14.00 (8.00, 15.00)14.00 (8.00, 15.00)< 0.001Severe sepsis (%)56 (16.5%)486 (13.6%)516 (11.3%)701 (13.0%)0.002 Interventions (1st 24 h) (%) MV use189 (55.8%)2052 (57.3%)2777 (60.8%)3488 (64.8%)< 0.001Vasoactive agent use189 (55.8%)2181 (60.9%)2985 (65.4%)3610 (67.1%)< 0.001ECMO use0 (0%)7 (0.2%)15 (0.3%)32 (0.6%)0.016CRRT use14 (4.1%)184 (5.1%)311 (6.8%)542 (10.1%)< 0.001 Comorbidities (%) Hypertension118 (34.8%)1419 (39.6%)2131 (46.7%)2777 (51.6%)< 0.001Diabetes39 (11.5%)584 (16.3%)971 (21.3%)1718 (31.9%)< 0.001MI52 (15.3%)655 (18.3%)895 (19.6%)1033 (19.2%)0.146CHF90 (26.5%)1033 (28.8%)1275 (27.9%)1630 (30.3%)0.047Cerebrovascular59 (17.4%)593 (16.6%)666 (14.6%)709 (13.2%)< 0.001COPD126 (37.2%)901 (25.2%)1032 (22.6%)1542 (28.7%)< 0.001Liver45 (13.3%)482 (13.5%)582 (12.8%)802 (14.9%)0.017Renal66 (19.5%)704 (19.7%)863 (18.9%)1106 (20.6%)0.230Malignancy63 (18.6%)460 (12.8%)525 (11.5%)503 (9.4%)< 0.001AIDS5 (1.5%)35 (1.0%)26 (0.6%)10 (0.2%)< 0.001 Vital signs (1st 24 h) HR (beats/min)85.83 (77.71, 97.12)84.58 (75.39, 96.53)84.12 (75.52, 95.10)84.66 (76.42, 96.02)0.043RR (beats/min)19.10 (16.64, 22.20)18.50 (16.33, 21.58)18.37 (16.36, 21.36)19.02 (16.96, 21.79)< 0.001MAP (mmHg)74.71 (69.21, 82.16)75.17 (69.89, 81.74)75.29 (70.36, 81.27)75.03 (70.19, 81.02)0.694Temperature (℃)36.78 (36.46, 37.03)36.82 (36.53, 37.19)36.84 (36.57, 37.20)36.90 (36.62, 37.27)< 0.001 Laboratory tests (1st 24 h) WBC (109/L)13.20 (9.80, 17.95)13.90 (9.90, 18.70)14.30 (10.60, 18.80)15.20 (11.30, 19.90)< 0.001HCT (%)28.60 (24.60, 33.15)28.50 (24.50, 33.00)28.60 (24.50, 33.10)29.10 (25.20, 33.90)< 0.001HGB (g/dL)9.20 (8.00, 10.80)9.50 (8.20, 11.0)9.60 (8.20, 11.10)9.70 (8.40, 11.30)< 0.001PLT (109/L)197.00 (124.00, 294.50)149.00 (105.00, 218.00)144.00 (104.00, 201.00)156.00 (113.00, 212.00)< 0.001PT (s)14.00 (12.70, 17.00)15.00 (13.10, 18.00)15.20 (13.50, 17.90)15.00 (13.50, 17.60)< 0.001PTT (s)33.50 (29.05, 44.65)34.40 (28.90, 46.00)34.00 (29.10, 46.50)33.10 (28.50, 43.90)< 0.001INR1.20 (1.10, 1.40)1.20 (1.10, 1.40)1.20 (1.10, 1.40)1.20 (1.10, 1.40)0.071Sodium (mmol/L)140.00 (138.00, 144.00)140.00 (137.00, 143.00)140.00 (138.00, 143.00)140.00 (138.00, 142.00)0.700Potassium (mmol/L)4.50 (4.10, 5.00)4.40 (4.10, 4.90)4.50 (4.20, 4.90)4.60 (4.20, 5.00)< 0.001Calcium (mmol/L)8.50 (8.00, 9.10)8.40 (8.00, 8.90)8.40 (8.00, 8.90)8.40 (8.00, 8.90)0.143Bicarbonate (mmol/L)24.00 (21.00, 27.00)24.00 (22.00, 26.00)24.00 (22.00, 26.00)24.00 (22.00, 27.00)< 0.001Anion gap (mmol/L)16.00 (14.00, 19.00)16.00 (13.00, 19.00)15.00 (13.00, 18.00)15.00 (13.00, 18.00)< 0.001Glu (mg/dl)125.00 (107.00, 146.50)126.19 (111.00, 149.50)130.26 (117.35, 154.00)135.33 (121.00, 162.28)< 0.001Cr (mg/dL)1.00 (0.70, 1.60)1.00 (0.80, 1.60)1.10 (0.80, 1.60)1.10 (0.90, 1.80)< 0.001BUN (mmol/L)24.00 (15.00, 39.00)21.00 (14.00, 34.00)21.00 (15.00, 33.00)22.00 (16.00, 36.00)< 0.001Urine output (mL)1150.00 (773.50, 1890.0)1560.0 (945.75, 2403.8)1,650.00 (1,030.00, 2,455.00)1,610.00 (1,000.00, 2,380.00)< 0.001In-hospital death (%)92 (27.1%)639 (17.8%)645 (14.1%)753 (14.0%)< 0.00190-day death (%)141 (41.6%)1029 (28.7%)970 (21.3%)1,039 (19.3%)< 0.001Sequential Organ Failure Assessment: SOFA, SAPS II: Simplified Acute Physiology Score II, Glasgow Coma Scale: GCS, MV: mechanical ventilation, ECMO: extracorporeal membrane oxygenation, CRRT: continuous renal replacement therapy, MI: myocardial infarction, CHF: congestive heart failure, COPD: chronic obstructive pulmonary disease, AIDS: acquired immune deficiency syndrome, HR: heart rate, RR: respiratory rate, MAP: mean arterial pressure, WBC: white blood cell count, HCT: haematocrit, HGB: haemoglobin, PLT: platelet counts, PT: prothrombin time, PTT: partial thrombin time, INR: international normalized ratio, AG: anion gap, Glu: blood glucose, Cr: creatinine, BUN: blood urea nitrogen. Baseline characteristics of subjects in the different BMI levels group. Sequential Organ Failure Assessment: SOFA, SAPS II: Simplified Acute Physiology Score II, Glasgow Coma Scale: GCS, MV: mechanical ventilation, ECMO: extracorporeal membrane oxygenation, CRRT: continuous renal replacement therapy, MI: myocardial infarction, CHF: congestive heart failure, COPD: chronic obstructive pulmonary disease, AIDS: acquired immune deficiency syndrome, HR: heart rate, RR: respiratory rate, MAP: mean arterial pressure, WBC: white blood cell count, HCT: haematocrit, HGB: haemoglobin, PLT: platelet counts, PT: prothrombin time, PTT: partial thrombin time, INR: international normalized ratio, AG: anion gap, Glu: blood glucose, Cr: creatinine, BUN: blood urea nitrogen. Association of BMI with in-hospital mortality Overall, the in-hospital mortality following ICU admission was 15.4%. In the analysis, patients in the obesity group had significantly lower in-hospital mortality (14.0%; 753 of 5379), compared to those in the overweight group (14.1%; 645 of 4564), normal weight group (17.8%; 639 of 3582) and the low weight group (27.1%; 92 of 339, P < 0.001, Table 1). Table S2 displays the variables associated with in-hospital mortality in patients with sepsis (P < 0.05). We utilized adjusted multivariable logistic regression models to examine the relationship between BMI and in-hospital mortality, as shown in Table 2. Using the normal weight group as the reference, the adjusted odds ratios (ORs) for in-hospital mortality in Model 1 for the low weight, overweight, and obesity groups were 1.72 (95% confidence interval [CI]: 1.32–2.20), 0.76 (95% CI: 0.67–0.85), and 0.75 (95% CI: 0.67–0.84), respectively. In Model 2, after adjusting for age and gender, the corresponding ORs compared with the normal weight group were 1.65 (95% CI: 1.27–2.12), 0.78 (95% CI: 0.69–0.88), and 0.79 (95% CI: 0.70–0.89), respectively. In Model 3, after further adjustment for additional covariates including comorbidities and interventions, the ORs compared with the normal weight group were 1.68 (95% CI: 1.29–2.20), 0.77 (95% CI: 0.67–0.88), and 0.74 (95% CI: 0.65–0.85), respectively. The results from the different models were consistent, with the P trend for BMI categories showing a statistically significant association with in-hospital mortality (P trend < 0.001). This suggests that higher BMI levels were associated with lower in-hospital mortality, and patients in the low weight group had the highest odds of mortality. Table 2Logistic model of the effect of different levels of BMI on in-hospital mortality.GroupsβSEWaldOR (95%CI) P P for trendModel 1a< 0.001Low weight group0.540.13017.311.72 (1.32 ~ 2.20)< 0.001Normal weight group1.00Overweight group-0.280.06120.690.76 (0.67 ~ 0.85)< 0.001Obesity group-0.290.05924.070.75 (0.67 ~ 0.84)< 0.001Model 2b< 0.001Low weight group0.500.13114.511.65 (1.27 ~ 2.12)Normal weight group1.00Overweight group-0.250.06116.990.78 (0.69 ~ 0.88)< 0.001Obesity group-0.230.05915.580.79 (0.70 ~ 0.89)< 0.001Model 3c< 0.001Low weight group0.520.12016.441.68 (1.29 ~ 2.20)< 0.001Normal weight group1.00Overweight group-0.270.07114.120.77 (0.67 ~ 0.88)< 0.001Obesity group-0.300.07117.630.74 (0.65 ~ 0.85)< 0.001aUnadjusted.bAdjusted for age, and gender.cAdjusted for age, gender, married, insurance, admission type, SOFA score, SAPS II score, GCS, severe sepsis, MV use, vasoactive agent use, ECMO use, CRRT use, hypertension, MI, CHF, cerebrovascular, COPD, liver disease, malignancy, HR, RR, MAP, temperature, HCT, PLT, PT, PTT, INR, sodium, potassium, calcium, bicarbonate, AG, Glu, Cr, BUN and Urine output. Logistic model of the effect of different levels of BMI on in-hospital mortality. aUnadjusted. bAdjusted for age, and gender. cAdjusted for age, gender, married, insurance, admission type, SOFA score, SAPS II score, GCS, severe sepsis, MV use, vasoactive agent use, ECMO use, CRRT use, hypertension, MI, CHF, cerebrovascular, COPD, liver disease, malignancy, HR, RR, MAP, temperature, HCT, PLT, PT, PTT, INR, sodium, potassium, calcium, bicarbonate, AG, Glu, Cr, BUN and Urine output. Association of BMI with 90-day mortality Overall, the 90-day mortality following ICU admission was 22.9%. As seen in Fig. 2, mortality increased rapidly during the first four weeks following ICU admission and continued to rise steadily thereafter. Crude 90-day mortality was significantly higher among patients in the low weight group (41.6%; 141 of 339), compared with patients in the normal weight group (28.7%; 1029 of 3582), overweight group (21.3%; 970 of 4564), and the obesity group (19.3%; 1039 of 5379; P < 0.001, Table 1). Table S3 shows the variables associated with 90-day mortality in patients with sepsis (P < 0.05). We utilized multivariable Cox proportional hazards models to examine the relationship between BMI and 90-day mortality, as shown in Table S4. Without adjusting for confounding, the unadjusted hazard ratios (HRs) for 90-day mortality in participants in the low weight, overweight, and obesity groups, compared with those in the normal weight group, were 1.61 (95% CI: 1.35–1.92), 0.72 (95% CI: 0.66–0.78), and 0.64 (95% CI: 0.59–0.70), respectively, in Model 1 (P < 0.001). In Model 2, after adjusting for age and sex, the HRs for 90-day mortality in participants in the low weight, overweight, and obesity groups were 1.57 (95% CI: 1.31–1.87), 0.74 (95% CI: 0.67–0.80), and 0.69 (95% CI: 0.64–0.76), respectively, compared with the normal weight group (P < 0.001). In Model 3, after further adjustment for additional factors such as comorbidities and interventions, the HRs for 90-day mortality in participants in the low weight, overweight, and obesity groups were 1.62 (95% CI: 1.34–1.94), 0.78 (95% CI: 0.71–0.85), and 0.74 (95% CI: 0.68–0.81), respectively, compared with the normal weight group (P < 0.001). These results suggest that higher BMI levels were associated with lower 90-day mortality across all models, with the low weight group having the highest relative risk for 90-day mortality. K-M curves were used to compare the 90-day mortality of sepsis groups with different BMI levels. Threshold effect analysis When BMI was analyzed as a continuous variable, an L-shaped association was observed between BMI and 90-day mortality. The adjusted odds of 90-day mortality were higher at lower BMI values, reached the lowest level at approximately 32 kg/m², and then plateaued or increased slightly at higher BMI values (Fig. 3). In addition, Cox proportional hazards regression with restricted cubic spline functions showed a similar nonlinear association between BMI and 90-day mortality (Figure S1). After adjustment for age, sex, marital status, insurance status, admission type, SOFA score, SAPS II score, GCS, severe sepsis, interventions, comorbidities, vital signs, and laboratory variables, higher BMI remained nonlinearly associated with lower in-hospital and 90-day mortality within the observed range. Association between BMI and the adjusted risk of 90-day mortality in patients with sepsis. Subgroup analyses Subgroup analyses for the correlation between different BMI levels and mortality were presented in Tables 3 and S5. The participants were divided into subgroups according to age and sex. The results showed that the association between BMI categories and in-hospital mortality remained generally consistent across age and sex subgroups (Tables 3, S5, P < 0.05). Table 3Logistic model of the effect of BMI on in-hospital mortality at different age and sex levels.GroupsβSEWaldOR (95%CI) P P for trend<65 years< 0.001Low weight group0.570.13518.721.77 (1.30 ~ 2.32)< 0.001Normal weight group1.00Overweight group-0.320.1207.160.73 (0.57 ~ 0.92)0.007Obesity group-0.260.1145.240.77 (0.62 ~ 0.96)0.022≥ 65 years< 0.001Low weight group0.500.12515.901.65 (1.26 ~ 2.16)< 0.001Normal weight group1.00Overweight group-0.220.0896.170.80 (0.67 ~ 0.95)0.013Obesity group-0.280.0939.020.76 (0.63 ~ 0.91)0.003Male< 0.001Low weight group0.550.13017.841.73 (1.32 ~ 2.24)< 0.001Normal weight group1.00Overweight group-0.440.09222.370.65 (0.54 ~ 0.77)< 0.001Obesity group-0.520.09629.790.59 (0.49 ~ 0.71)< 0.001Female0.308Low weight group0.340.2022.771.40 (0.94 ~ 2.07)0.096Normal weight group1.00Overweight group-0.020.1110.030.98 (0.79 ~ 1.22)0.867Obesity group-0.020.1070.050.98 (0.79 ~ 1.20)0.828 Logistic model of the effect of BMI on in-hospital mortality at different age and sex levels. External cohort analysis In the external dataset, 498 patients with sepsis were included and categorized into two groups according to the BMI threshold identified from the primary cohort: the non-high BMI threshold group (n = 450) and the high BMI threshold group (n = 48, BMI ≥ 32 kg/m²). Before adjustment, baseline covariates were imbalanced between the two groups. After IPTW, covariate balance improved substantially, with most SMDs below 0.1 (Table S6 and Figure S2). In the weighted cohort, in-hospital mortality remained significantly higher in the non-high BMI threshold group than in the high BMI threshold group [10.9 (23.6%) vs. 3.7 (7.9%), P = 0.036]. Weighted logistic regression showed that a high BMI threshold was associated with lower in-hospital mortality (ORs = 0.28, 95% CI 0.09–0.83, P = 0.022). Kaplan-Meier analysis also showed significantly higher 90-day mortality in the non-high BMI threshold group than in the high BMI threshold group (Figure S3, P < 0.001). As a sensitivity analysis, we additionally performed propensity score matching. After 1:2 matching, 96 patients in the non-high BMI threshold group were matched with 48 patients in the high BMI threshold group, and baseline covariates were adequately balanced (Table S7 and Figure S4). In the matched cohort, the association between high BMI threshold and lower in-hospital mortality remained significant and consistent (ORs = 0.32, 95% CI 0.14–0.71, P = 0.005). Similarly, 90-day mortality was lower in the high BMI threshold group, with results generally consistent with the IPTW analysis (Figure S5; HRs = 0.40, 95% CI 0.23–0.71, P = 0.001). Discussion In this study, obesity was common among ICU patients with sepsis, with approximately one-third of the cohort classified as obese. We found that BMI, whether analyzed as a continuous or categorical variable, was inversely associated with in-hospital and 90-day mortality after adjustment for measured covariates. Similar patterns were observed in subgroup analyses, although some subgroup-specific associations were not statistically significant. BMI is a widely used anthropometric index defined as weight divided by height squared and is recommended by the World Health Organization (WHO) for general risk stratification7,8. According to WHO criteria, obesity is defined as a BMI ≥ 30 kg/m²18,19. In the present study, higher BMI was associated with lower short-term mortality among ICU patients with sepsis after adjustment for measured confounders. This pattern is broadly consistent with prior observational studies. For example, Prescott et al. reported an association between higher BMI and lower 1-year mortality in patients with severe sepsis20, and meta-analyses have similarly described lower adjusted mortality among overweight or obese ICU patients13. However, findings across studies have been heterogeneous, which may partly reflect differences in study design, patient populations, and adjustment for confounding variables. Several mechanisms have been proposed in prior literature to explain the observed differences in outcomes across BMI categories in critically ill populations. One commonly discussed explanation is that higher BMI may reflect greater metabolic reserve, which could be advantageous during the hypercatabolic state of sepsis, where energy expenditure and protein breakdown are markedly increased21,22. In this context, patients with higher energy and nutrient reserves may better tolerate prolonged critical illness and recovery phases23,24. In addition, differences in inflammatory and immune responses across BMI categories have been suggested. Adipose tissue is not merely an energy storage organ but also an active endocrine organ that secretes adipokines and cytokines involved in immune regulation. Altered levels of leptin, adiponectin, and other inflammatory mediators in individuals with higher BMI may modulate the host response to infection, potentially influencing disease progression and outcomes25. Another proposed explanation relates to lipid metabolism. Circulating lipoproteins may bind and neutralize bacterial endotoxins such as lipopolysaccharide, potentially attenuating the systemic inflammatory response in sepsis26. This mechanism has been hypothesized to contribute to differential survival patterns observed across BMI strata. Furthermore, patients with higher BMI may have differences in baseline nutritional status and muscle mass distribution27. Greater absolute skeletal muscle mass may provide a physiological buffer against the severe protein catabolism observed in sepsis, thereby supporting organ function during critical illness28,29. Taken together, these hypotheses suggest that the inverse association between BMI and mortality observed in this study is likely multifactorial and reflects complex interactions among metabolic reserve, immune regulation, and host response to critical illness rather than a single causal pathway. Despite the importance of these findings, several limitations should be acknowledged. First, this retrospective ICU-based study is inherently subject to selection bias and residual confounding. Exclusion of patients without BMI data may have introduced bias if these patients differed in illness severity or clinical characteristics. Second, although multiple severity indicators were adjusted for, residual confounding by unmeasured factors remains likely. Key variables such as nutritional status, albumin levels, sarcopenia, frailty, pre-ICU weight loss, and fluid balance were unavailable or incompletely recorded, and may influence both BMI classification and outcomes. Third, BMI is an imperfect surrogate for body composition in critically ill patients, as it does not distinguish fat from lean mass or capture conditions such as sarcopenic obesity. In addition, BMI measurements in the ICU may be affected by fluid resuscitation, edema, or estimation of body weight, leading to potential misclassification. Finally, the external cohort analysis should be interpreted as exploratory and supportive rather than confirmatory, given its relatively small sample size and limited number of patients in the high BMI category. Although supplementary analyses such as propensity score matching were performed to improve covariate balance, these limitations restrict the strength of inference that can be drawn. In conclusion, in this retrospective cohort of ICU patients with sepsis, higher BMI was associated with lower in-hospital and 90-day mortality compared with low or normal BMI after adjustment. These findings are observational and associative in nature and should not be interpreted as demonstrating a protective effect or a causal relationship. Given the possibility of residual confounding, selection bias, and database-related limitations, the results should be interpreted cautiously. Further studies in larger and more diverse cohorts are needed to examine the reproducibility and potential mechanisms underlying these associations. Electronic Supplementary Material Below is the link to the electronic supplementary material. Supplementary Material 1 Author contributions Yunya Zhu and Wenyuan Zhang contributed equally to this work. Ying Li conceptualized the research aims, planned the analyses and guided the literature review. Wenyuan Zhang extracted the data from the MIMIC-IV database. Yunya Zhu and Wenyuan Zhang participated in processing the data and doing the statistical analysis. Yunya Zhu, Yuting Zhong and Luwen Liang wrote the first draft of the paper and the other authors provided comments and approved the final manuscript. The author(s) read and approved the final manuscript. Funding The authors declare no funding involved. Data availability The datasets given in this investigation are accessible through the MIMIC-IV database (https://www.physionet.org/content/mimiciv/3.1/). Code Availability Not applicable. Declarations Consent for publication Not applicable. Competing interests The authors declare no competing interests. Ethics approval This retrospective analysis conducted at the First Affiliated Hospital of Gannan Medical University was approved by the Ethical Committee (No. LLSC-2025-503). This study was conducted in accordance with the ethical guidelines set forth in the Declaration of Helsinki. Additionally, the use of the MIMIC-IV database was approved by both the Massachusetts Institute of Technology and the Beth Israel Deaconess Medical Center. The datasets utilized in this research can be accessed through the MIMIC database. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Yunya Zhu and Wenyuan Zhang contributed equally References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The datasets given in this investigation are accessible through the MIMIC-IV database (https://www.physionet.org/content/mimiciv/3.1/). Not applicable.