Association of prognostic nutritional index with mortality in patients with sepsis-associated delirium: a retrospective MIMIC-IV study Authors: Shengnan Kong, Chunling Yang, Xinya Li, Qin Lai, Jiafang Zhang, Yu Wang, Jun Lyu Published: 2026 ABSTRACT AbstractBackgroundA concerning sign of severe brain dysfunction, sepsis-associated delirium (SAD) is associated with poor clinical outcomes. The prognostic value of the Prognostic Nutritional Index (PNI) for patients with SAD is unclear.ObjectiveTo determine the association between PNI and all-cause mortality at 90, 180, and 360 days in SAD patients.MethodsWe classified SAD patients into low-PNI and high-PNI groups based on PNI. We then used Kaplan-Meier (KM) survival analysis, multivariate Cox regression, restricted cubic spline (RCS) models, inverse probability of treatment weighting (IPTW), and subgroup analysis to evaluate the association between these groups and all-cause mortality at 90, 180, and 360 days. A receiver operating characteristic (ROC) curve analysis was performed to evaluate the predictive ability of PNI.ResultsAmong 2,160 patients, the low PNI group had significantly higher mortality than the high PNI group at 90 days (51.9% vs. 33.5%), 180 days (56.4% vs. 37.1%), and 360 days (60.6% vs. 41.5%). After adjustment, Cox regression showed low PNI was associated with a 1.55-fold higher 90-day mortality (95% CI: 1.34–1.79, p < 0.001). A restricted cubic spline model revealed a nonlinear PNI-outcome relationship. These results were robust in IPTW and subgroup analyses, and ROC analysis demonstrated predictive performance comparable to that of SOFA and albumin.ConclusionThe PNI is a simple, useful, and easily accessible tool that can help identify patients with SAD who are at increased risk of adverse outcomes at an early stage.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12883-026-05115-w.Keywords: PNI, Patients with sepsis-associated delirium, All-cause mortality, Nutritional status, Risk prediction Introduction High death rates are linked to sepsis, a serious systemic illness that results from an improper immunological response to infection [1]. It commonly presents with acute organ dysfunction. According to current epidemiological data, approximately 19 million people worldwide experience sepsis annually, which leads to about 5 million fatalities. Low- and middle-income nations have the highest death rates [2]. This abnormal metabolic state rapidly depletes nutritional reserves, resulting in hypoalbuminaemia, and also severely impairs immune function. Persistent hypercatabolism and loss of lean body mass are common side effects of sepsis. The consequences of suboptimal nutrition in the ICU, particularly low protein delivery, have a long-lasting impact on survivors’ well-being and increase the risk of death later in life [3]. Furthermore, malnutrition may persist after ICU discharge in some patients, prolonging recovery, increasing the likelihood of functional impairment and further compromising quality of life [4]. Delirium is a complex clinical syndrome characterised by acute brain dysfunction, with abrupt and fluctuating impairments in consciousness, attention and cognitive function [5]. Delirium that occurs alongside sepsis is known as sepsis-associated delirium (SAD) [6]. SAD is a core feature of sepsis-associated encephalopathy. It typically follows an insidious yet rapidly progressive course, posing significant management challenges once fully manifest. This underlines the critical importance of early recognition and timely treatment [7]. From a clinical perspective, it is therefore essential to identify simple and reliable markers to predict or evaluate the impact of SAD on patient outcomes. There is growing evidence that malnutrition and low nutritional status are associated with the development and course of delirium, particularly in elderly and critical care patients [8–12]. The Geriatric Nutritional Risk Index (GNRI) [13, 14] and the Mini Nutritional Assessment–Short Form (MNA-SF) [15] are two nutritional assessment tools that have been linked to delirium in several clinical settings, including surgical and critical care units. Dehydration and electrolyte abnormalities brought on by malnutrition can impact neurotransmitter function and neuronal excitability. Furthermore, nutritional disruptions can directly affect cognitive processes like attention and consciousness, making patients more susceptible to delirium due to the brain’s high metabolic demands [16]. Furthermore, it has been demonstrated that systemic inflammation worsens neuroinflammation and weakens the blood-brain barrier [17]. The pathophysiology of sepsis-associated delirium (SAD) revolves around these key mechanisms. Similarly, the Prognostic Nutritional Index (PNI), which reflects nutritional and immune status, has been reported to correlate with delirium in specific patient groups, including older adults with hip fractures [18]. Taken together, these findings suggest that nutritional assessment tools capture clinically relevant vulnerability in relation to delirium risk. However, it is unclear whether PNI retains its prognostic relevance in patients with SAD, and further investigation is warranted in this specific high-risk population. Patients with sepsis often have low levels of serum albumin, an acute-phase protein that has been consistently linked to poorer clinical outcomes [19]. Sepsis also causes severe immune dysregulation, which frequently shows up as lymphopenia, an immunosuppressive condition that is highly linked to unfavorable outcomes [20]. High natural killer (NK) cell counts in patients with severe sepsis are independently linked to higher mortality, according to recent studies [21]. Peripheral lymphocyte counts and serum albumin are both included in the PNI, a composite nutritional assessment. Lower PNI values are consistently linked to unfavorable outcomes in a variety of clinical conditions, according to mounting evidence. Previous studies have suggested that lower PNI levels have been linked to greater mortality across diverse critically ill populations. Similar correlations have also been documented in specific disease contexts, including sepsis [22], intestinal obstruction [23], and severe ventilator-associated pneumonia [24]. However, these investigations primarily focused on short-term outcomes or mixed ICU populations. Furthermore, a comparative study of ICU patients found that, of four nutritional assessment tools, the PNI demonstrated the strongest independent predictive value for delirium, suggesting its clinical utility in evaluating delirium risk from a combined nutritional and immunological perspective [25]. However, the ability of PNI to predict mid- to long-term outcomes in patients with SAD has not yet been investigated. Our aim was therefore to examine the association between PNI and all-cause mortality at 90, 180, and 360 days. Materials and methods Data source The retrospective cohort design employed in this study is a common approach in clinical prognostic research [26]. The observational cohort study was conducted using data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database [27, 28]. De-identified clinical records for about 364,627 patients who underwent emergency or critical care at Boston’s Beth Israel Deaconess Medical Center between 2008 and 2022 are included in MIMIC-IV version 3.1 [29]. Since the database ensures patient confidentiality, individual informed permission is not necessary. After completing the required training and certification processes, access is allowed [27]. Study population This study examined adult patients with sepsis-associated delirium (SAD) who were admitted to the ICU for the first time. Sepsis was defined according to the Sepsis-3 criteria as suspected or confirmed infection accompanied by an increase in the SOFA score of at least two points [1]. Specifically, all suspected infection episodes were identified by the timestamp of antibiotic administration or blood culture, whichever occurred first, within a specific timeframe. If the antibiotic was administered first, culture sampling had to be completed within 24 h. If the blood culture was obtained first, the antibiotic had to be ordered within 72 h. Sepsis onset time (tsepsis) was defined as the moment when a suspected infection episode was associated with a SOFA score increase of ≥ 2 points, evaluated within a time window ranging from 48 h before to 24 h after the suspected infection time (tsuspicion) [30]. The diagnosis of delirium employed a tiered approach, with structured assessments serving as the primary method and unstructured clinical records used to aid in identification. Specifically, the Confusion Assessment Method for the ICU (CAM-ICU) was primarily used to evaluate and screen patients [31, 32]. This method consisted of two steps: First, a preliminary assessment was conducted using the Richmond Agitation–Sedation Scale (RASS); then, for patients with a RASS score of ≥ -3, the CAM-ICU delirium assessment was performed: (1) rapid changes or fluctuations in mental status; (2) inattention; (3) altered level of consciousness; (4) disorganized thinking. If a patient meets criteria 1 and 2 and satisfies either criterion 3 or 4, a diagnosis of delirium can be made. To reduce the potential for missed delirium cases in routine clinical practice due to the lack of formal assessment or documentation via the structured CAM-ICU assessment, nursing notes served as an adjunctive diagnostic method by identifying potential delirium cases through the recognition of key terms such as “delirium,” “confusion,” “agitation,” and “altered mental status” [33–35]. This supplementary strategy was not meant to replace the CAM-ICU assessment. Rather, it was intended to improve the sensitivity of delirium detection within the limitations of retrospective electronic medical record data. We excluded patients who met any of the following criteria: (1) ICU length of stay < 1 day, (2) No documented delirium assessment, (3) Patients with missing albumin or lymphocyte data, (4) Delirium onset before PNI measurement, and (5) Inconsistent temporal records. We analyzed 2,160 patients. Figure 1 presents the procedure used to select the study population. Inclusion and exclusion flowchart of the study. Abbreviations: ICU: Intensive Care Unit; MIMIC-IV: Medical Information Mart for Intensive Care IV Data extraction and variable selection Data from electronic health records were extracted using Structured Query Language (SQL) in this study [36]. The extracted data included the following: demographic characteristics; comorbidities; laboratory parameters; disease severity scores; treatments; medications; vital signs; and outcomes. All laboratory measurements were completed during the first twenty-four hours after admission to the ICU. The proportion of missing data did not exceed 15% for any variable. The ‘MICE’ package in R was used to build the random forest approach for multiple imputation of parameters with fewer than 15% missing data [37]. Outcomes and measures The primary outcome was all-cause mortality at 90 days following ICU admission. Secondary outcomes were all-cause mortality at 180 and 360 days. Index time was defined as the first available albumin and lymphocyte measurement within 24 h of ICU admission. Mortality follow-up began after this measurement. Statistical analysis This study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) standard [38]. Participants were divided into two groups based on PNI. PNI = 10 × serum albumin (g/dL) + 5 × total lymphocyte count(10⁹/L) [23]. In order to summarize the general characteristics of participants, descriptive statistics were used. Shapiro–Wilk test showed that most continuous variables were not normally distributed. When presenting data, categorical variables were presented as frequencies and percentages, while continuous variables were expressed as medians with interquartile ranges. Kaplan-Meier survival curves were used to analyze the relationship between PNI and 90-day total mortality across different, and the log-rank test was used to assess any significant differences between groups. To further explore the relationship between PNI and clinical outcomes, we employed Cox proportional hazards regression. For subsequent analyses, patients were categorized into low and high PNI groups according to PNI distribution. Progressively adjusted Cox regression models were constructed. The unadjusted model included only PNI. Model 1 was adjusted for demographic variables, including age and race. Model 2 was further adjusted for disease severity indicators, comorbidities, and laboratory parameters, including Sequential Organ Failure Assessment (SOFA), congestive heart failure, peripheral vascular disease, mild liver disease, alkaline phosphatase (ALP), white blood cell count (WBC), creatinine, red blood cell count (RBC), total bilirubin, and sodium. Model 3 additionally included vital signs and treatment-related variables, including oxygen saturation (SpO2), systolic blood pressure (SBP), diastolic blood pressure (DBP), temperature, and renal replacement therapy (RRT). Model 3 was considered the primary adjusted model for all subsequent analyses. Multicollinearity between variables was evaluated using variance inflation factors (VIFs); All VIFs were below 5, demonstrating there was no significant collinearity. The proportional hazards assumption was evaluated using Schoenfeld residuals, and no significant violation was observed for the overall models or for PNI. We turned to inverse probability of treatment weighting (IPTW) to account for any potential confounding variables to balance baseline covariates between the low-PNI and high-PNI groups, while standardized mean differences (SMDs) helped us gauge how well our covariates were balanced across groups. Any SMD values under 0.1 were deemed to represent sufficient equilibrium between the comparison groups [39]. To reduce the influence of extreme weights, IPTW weights were truncated at the 1st and 99th percentiles. Subsequently, the truncated weights were normalized within each treatment group to stabilize the weight distribution and maintain the effective weighted sample size comparable to the original sample size in each group. Restricted cubic splines (RCS) were used to include continuous PNI data into Cox regression models to examine the possibility of a nonlinear association between PNI and mortality. This method is well-suited to evaluating nonlinear relationships between continuous predictors and outcomes [40]. PNI was also standardized for additional RCS analysis. A receiver operating characteristic (ROC) curve analysis was performed to compare the predictive performance of PNI, albumin, and the SOFA in predicting all-cause mortality at 90, 180, and 360 days. Using subgroup analysis and statistical testing of interaction effects, the robustness of the association between PNI and mortality by age, sex, race, SOFA, APS III, and RRT status was assessed. The statistical analyses were conducted using R (version 4.5.1). Ethical and institutional approvals The Medical Information Mart for Intensive Care IV (MIMIC-IV) database was supported by the National Institute of Biomedical Imaging and Bioengineering of the National Institutes of Health under grants R01-EB001659 (2003–2013) and R01-EB017205 (2014–2018). Approval for the database was granted by the Institutional Review Boards of Beth Israel Deaconess Medical Center (Boston, Massachusetts) and the Massachusetts Institute of Technology (Cambridge, Massachusetts). Results Characteristics of included patients The final study cohort comprised 2,160 patients, as summarised in Table 1. The median PNI was 30.00 [25.01, 34.01], and the median age was 65.50 [54.34, 76.06] years. Males accounted for 59.2% of the cohort. Patients were divided into low- and high-PNI groups based on their PNI levels. Patients in the low-PNI group had longer ICU and total hospital stays, poorer physiological condition, more severe illness, and received more therapeutic interventions than those in the high-PNI group. Baseline characteristics for the 90-day survival and mortality cohorts are detailed in Supplementary Table 1. We compared the baseline characteristics of the included and excluded patients to evaluate potential selection bias. The results are presented in Supplementary Table 2. Table 1Baseline Characteristics of the Study PopulationVariableOverall(n = 2,160)Low PNI group(n = 720)High PNI group(n = 1,440)P valuePNI30.00 [25.01, 34.01]23.01 [21.00, 25.01]33.00 [30.00, 36.01]< 0.001 Personal characteristics Age (years old)65.50 [54.34, 76.06]65.03 [54.58, 75.46]65.78 [54.32, 76.43]0.462Sex (%)0.016Male1279 (59.2)400 (55.6)879 (61.0)Female881 (40.8)320 (44.4)561 (39.0)Race (%)0.146White1199 (55.5)416 (57.8)783 (54.4)Other961 (44.5)304 (42.2)657 (45.6)Hospital LOS (days)13.75 [7.78, 22.83]15.74 [8.33, 25.37]13.00 [7.61, 22.22]< 0.001ICU LOS (days)6.06 [3.34, 11.47]6.43 [3.52, 12.59]5.88 [3.23, 10.82]0.023Weight (Kg)81.00 [67.47, 97.43]77.80 [64.60, 95.00]82.50 [69.00, 99.30]< 0.001 Scores GCS14.00 [10.00, 15.00]14.00 [9.00, 15.00]14.00 [10.00, 15.00]0.147APSIII59.00 [46.00, 77.00]68.00 [53.00, 84.00]55.00 [43.00, 72.25]< 0.001SOFA3.00 [1.00, 5.00]3.00 [1.00, 5.00]2.00 [0.00, 5.00]< 0.001CCI5.00 [3.00, 8.00]6.00 [3.75, 8.00]5.00 [3.00, 8.00]0.063LODS7.00 [5.00, 9.00]8.00 [6.00, 10.00]7.00 [5.00, 9.00]< 0.001OASIS38.00 [32.00, 44.00]39.00 [34.00, 45.00]37.00 [32.00, 43.00]< 0.001SIRS3.00 [2.00, 4.00]3.00 [3.00, 4.00]3.00 [2.00, 4.00]0.007 Vital signs Heart rate (beats/minute)93.00 [80.00, 109.00]96.00 [83.00, 111.00]92.00 [78.00, 108.00]< 0.001Systolic Blood Pressure(mmHg)118.00 [102.00, 136.00]111.00 [98.00, 130.00]121.00 [104.00, 138.00]< 0.001Diastolic Blood Pressure(mmHg)66.00 [56.00, 78.25]63.75 [53.00, 74.00]68.00 [57.00, 80.00]< 0.001Mean Blood Pressure(mmHg)81.00 [69.00, 93.00]77.00 [66.75, 89.00]82.00 [71.00, 95.00]< 0.001Respiration rate (beats/minute)21.00 [18.00, 25.00]21.00 [18.00, 26.00]21.00 [17.00, 25.00]0.328Temperature (°C)36.83 [36.44, 37.22]36.78 [36.44, 37.17]36.89 [36.50, 37.28]0.001Spo2 (%)97.00 [94.00, 100.00]97.00 [94.00, 100.00]97.00 [94.00, 100.00]0.606 Laboratory tests WBC (109 /L)12.35 [8.20, 18.30]12.70 [8.10, 19.40]12.20 [8.20, 17.60]0.202Anion gap (mmol/L)16.00 [13.00, 19.00]15.00 [12.00, 18.00]16.00 [13.00, 20.00]< 0.001Total bilirubin (mg/dL )0.80 [0.40, 1.70]0.90 [0.50, 2.32]0.70 [0.40, 1.50]< 0.001Calcium (mg/dL)8.20 [7.60, 8.80]7.80 [7.20, 8.40]8.40 [7.90, 8.90]< 0.001Creatinine (mg/dL)1.40 [0.90, 2.40]1.40 [0.90, 2.30]1.30 [0.90, 2.40]0.522Glucose (mg/dL)138.00 [106.00, 190.00]129.00 [100.00, 186.00]143.00 [110.00, 192.00]< 0.001Sodium (mEq/L)138.00 [134.00, 142.00]138.00 [133.00, 141.00]138.00 [135.00, 142.00]0.013Potassium (mEq/L)4.25 [3.80, 4.80]4.20 [3.70, 4.80]4.30 [3.80, 4.90]0.001Hemoglobin (g/dL)10.30 [8.60, 12.20]9.40 [8.00, 10.90]10.90 [9.00, 12.70]< 0.001Platelet (109 /L)176.00 [111.00, 257.00]158.00 [96.00, 254.25]186.00 [122.00, 259.00]< 0.001RBC (10¹²/L)3.47 [2.87, 4.14]3.15 [2.65, 3.66]3.68 [3.01, 4.30]< 0.001Lymphocytes (109 /L)0.91 [0.52, 1.46]0.81 [0.43, 1.33]0.95 [0.57, 1.53]< 0.001Albumin (g/dL)3.00 [2.50, 3.40]2.30 [2.10, 2.50]3.30 [3.00, 3.60]< 0.001Monocyte count (K/uL)5.00 [3.10, 8.00]4.60 [2.60, 7.30]5.40 [3.60, 8.00]< 0.001INR1.40 [1.20, 1.80]1.50 [1.30, 2.00]1.30 [1.10, 1.70]< 0.001BUN (mg/dL)28.00 [17.00, 48.00]29.00 [17.00, 50.00]27.00 [17.00, 47.00]0.363PTT (s)32.45 [28.10, 41.20]34.80 [29.60, 45.68]31.40 [27.50, 39.50]< 0.001ALT (U/L)31.00 [17.00, 74.00]33.00 [17.00, 79.25]30.00 [17.00, 71.00]0.281ALP (U/L)88.00 [63.00, 133.00]104.00 [65.00, 166.25]84.00 [62.00, 120.00]< 0.001AST (U/L)51.00 [27.00, 128.25]55.00 [28.00, 157.50]49.00 [26.00, 116.00]0.025 Comorbidities Diabetes with cc (%)0.268Yes309 (14.3)94 (13.1)215 (14.9)No1851 (85.7)626 (86.9)1225 (85.1)Diabetes without cc (%)0.05Yes565 (26.2)169 (23.5)396 (27.5)No1595 (73.8)551 (76.5)1044 (72.5)Renal disease (%)0.07Yes604 (28.0)183 (25.4)421 (29.2)No1556 (72.0)537 (74.6)1019 (70.8)Myocardial infarct (%)0.007Yes419 (19.4)116 (16.1)303 (21.0)No1741 (80.6)604 (83.9)1137 (79.0)Congestive heart failure (%)0.002Yes760 (35.2)221 (30.7)539 (37.4)No1400 (64.8)499 (69.3)901 (62.6)Cerebrovascular disease (%)0.065Yes365 (16.9)106 (14.7)259 (18.0)No1795 (83.1)614 (85.3)1181 (82.0)Chronic pulmonary disease (%)0.236Yes579 (26.8)181 (25.1)398 (27.6)No1581 (73.2)539 (74.9)1042 (72.4)Malignant cancer (%)< 0.001Yes305 (14.1)145 (20.1)160 (11.1)No1855 (85.9)575 (79.9)1280 (88.9)Mild liver disease (%)0.005Yes577 (26.7)220 (30.6)357 (24.8)No1583 (73.3)500 (69.4)1083 (75.2)Severe liver disease (%)0.005Yes317 (14.7)128 (17.8)189 (13.1)No1843 (85.3)592 (82.2)1251 (86.9)Rheumatic disease (%)0.465Yes88 (4.1)33 (4.6)55 (3.8)No2072 (95.9)687 (95.4)1385 (96.2)Peripheral vascular disease (%)0.396Yes239 (11.1)86 (11.9)153 (10.6)No1921 (88.9)634 (88.1)1287 (89.4)Peptic ulcer disease (%)0.002Yes87 (4.0)43 (6.0)44 (3.1)No2073 (96.0)677 (94.0)1396 (96.9)Dementia (%)0.406Yes141 (6.5)42 (5.8)99 (6.9)No2019 (93.5)678 (94.2)1341 (93.1) Drugs and treatments Mechanical ventilation (%)0.484Yes1721 (79.7)567 (78.8)1154 (80.1)No439 (20.3)153 (21.2)286 (19.9)RRT (%)0.002Yes502 (23.2)196 (27.2)306 (21.2)No1658 (76.8)524 (72.8)1134 (78.8)Vasoactive agent (%)< 0.001Yes1353 (62.6)505 (70.1)848 (58.9)No807 (37.4)215 (29.9)592 (41.1)Sedative administration (%)0.157Yes1951 (90.3)660 (91.7)1291 (89.7)No209 (9.7)60 (8.3)149 (10.3) Outcomes ICU mortality (%)< 0.001Yes436 (20.2)188 (26.1)248 (17.2)No1724 (79.8)532 (73.9)1192 (82.8)In-hospital mortality (%)< 0.001Yes605 (28.0)268 (37.2)337 (23.4)No1555 (72.0)452 (62.8)1103 (76.6)90-day all-cause mortality (%)< 0.001Yes856 (39.6)374 (51.9)482 (33.5)No1304 (60.4)346 (48.1)958 (66.5)180-day all-cause mortality (%)< 0.001Yes940 (43.5)406 (56.4)534 (37.1)No1220 (56.5)314 (43.6)906 (62.9)360-day all-cause mortality (%)< 0.001Yes1034 (47.9)436 (60.6)598 (41.5)No1126 (52.1)284 (39.4)842 (58.5)PNI Prognostic Nutritional Index, LOS length of stay, ICU intensive care unit, GCS Glasgow Coma Scale, APSIII Acute Physiology Score III, SOFA Sequential Organ Failure Assessment, CCI Charlson Comorbidity Index, LODS logistic organ dysfunction system, OASIS Oxford Acute Severity of Illness Score, SIRS Systemic Inflammatory Response Syndrome, SPO2 saturation of peripheral oxygen, WBC white blood cell count, RBC red blood cell count, INR, international normalized ratio, BUN blood urea nitrogen, PTT partial thromboplastin time, ALT alanine aminotransferase, ALP alkaline phosphatase, AST aspartate aminotransferase, RRT renal replacement therapyData are presented as median [interquartile range] for continuous variables and n (%) for categorical variables* A p-value < 0.05 was considered statistically significant Baseline Characteristics of the Study Population PNI Prognostic Nutritional Index, LOS length of stay, ICU intensive care unit, GCS Glasgow Coma Scale, APSIII Acute Physiology Score III, SOFA Sequential Organ Failure Assessment, CCI Charlson Comorbidity Index, LODS logistic organ dysfunction system, OASIS Oxford Acute Severity of Illness Score, SIRS Systemic Inflammatory Response Syndrome, SPO2 saturation of peripheral oxygen, WBC white blood cell count, RBC red blood cell count, INR, international normalized ratio, BUN blood urea nitrogen, PTT partial thromboplastin time, ALT alanine aminotransferase, ALP alkaline phosphatase, AST aspartate aminotransferase, RRT renal replacement therapy Data are presented as median [interquartile range] for continuous variables and n (%) for categorical variables * A p-value < 0.05 was considered statistically significant Primary Outcome The all-cause mortality rate within the entire study population over the 90-day period was 39.6% (Table 1). The 90-day mortality rate was significantly higher in the low PNI group than in the high PNI group (51.9% vs. 33.5%, p < 0.001). Kaplan-Meier survival analysis (Fig. 2) revealed that the probability of survival was significantly lower in the low PNI group than in the high PNI group during the 90-day follow-up period (log-rank test; p < 0.001). We used Cox proportional hazards regression to analyze the association of PNI on 90-day mortality. In all models, patients in the low PNI group were at significantly higher risk for 90-day mortality than those in the high PNI group (adjusted HR: 1.55, 95% CI: 1.34–1.79, p < 0.001). IPTW was performed to reduce potential confounding factors. The IPTW model included the following variables: demographic characteristics; disease severity indicators; comorbidities; laboratory parameters; vital signs; and treatment-related variables. These variables included age; race; SOFA; congestive heart failure; peripheral vascular disease; mild liver disease; ALP; WBC; creatinine; RBC; total bilirubin; sodium; SpO2; SBP; DBP; temperature; and RRT. After IPTW adjustment, the baseline covariates were well-balanced between the groups, with all standardized mean differences (SMDs) below 0.1. See Supplementary Tables 3 and Supplementary Fig. 1 for detailed balance diagnostics. The effective sample sizes (ESS) after weighting were 578 in the low-PNI group and 1,291 in the high-PNI group. IPTW was used to further reduce confounding, and the weighted cohort’s Cox regression results(adjusted HR: 1.48, 95% CI: 1.28–1.72, p < 0.001) agreed with the main analyses (Table 2). Restricted cubic spline (Fig. 3) analysis revealed a nonlinear association, which was observed in both the unadjusted model (P for overall < 0.001, P for non-linearity < 0.001) and the multivariable adjusted model (P for overall < 0.001, P for non-linearity < 0.001). The RCS model was fitted with four knots located at the 5th, 35th, 65th, and 95th percentiles of PNI, corresponding to PNI values of 19.00, 27.00, 32.01, and 41.00, respectively, with the reference point (HR = 1.0) set at 30.00 (the median PNI of the population). Consistent trends were observed in the RCS analysis of standardized PNI (Supplementary Fig. 3). The RCS model was fitted with four knots at the corresponding percentiles (values: -1.60, -0.42, 0.32, and 1.65) with the reference point set at 0. We performed ROC curve analysis to evaluate the predictive performance of PNI, albumin, and SOFA for 90-day mortality in patients with SAD. The area under the curve (AUC) was 0.599 (95% CI: 0.574–0.624) for PNI, 0.590 (95% CI: 0.566–0.614) for SOFA, and 0.597 (95% CI: 0.572–0.622) for albumin (Supplementary Fig. 4). These results indicate comparable discriminatory performance among the three indicators. Kaplan–Meier curves for 90-day mortality by PNI groups. Abbreviations: Kaplan-Meier survival curves for all-cause mortality stratified by PNI tertiles. Log-rank test P < 0.001 Table 2Cox regression analysis of the association between PNI and 90-day mortality in sepsis-associated delirium patientsGroupsUnivariable HR(95% CI)P valueModel 1 Adjusted HR(95% CI)P valueModel 2 Adjusted HR(95% CI)P valueModel 3 Adjusted HR(95% CI)P valueIPTW model HR(95% CI)P valueHigh PNI groupReferenceReferenceReferenceReferenceReferenceLow PNI group1.77 (1.55–2.03)< 0.0011.85 (1.61–2.11)< 0.0011.67 (1.45–1.92)< 0.0011.55 (1.34–1.79)< 0.0011.48 (1.28–1.72)< 0.001CI confidence interval, HR hazard ratio, SOFA Sequential Organ Failure Assessment, ALP alkaline phosphatase, WBC white blood cell count, RBC red blood cell count, SpO2 peripheral oxygen saturation, SBP systolic blood pressure, DBP diastolic blood pressure, RRT renal replacement therapy● Model 1 was adjusted for age and race● Model 2 was further adjusted based on Model 1 for severity scores (SOFA), comorbidities (congestive heart failure, peripheral vascular disease, mild liver disease), and laboratory parameters (ALP, WBC, creatinine, RBC, total bilirubin, and sodium)● Model 3 was further adjusted based on Model 2 for vital signs (SpO2, SBP, DBP, temperature) and treatment modalities (RRT). Model 3 was the primary adjusted model reported in the Results Cox regression analysis of the association between PNI and 90-day mortality in sepsis-associated delirium patients CI confidence interval, HR hazard ratio, SOFA Sequential Organ Failure Assessment, ALP alkaline phosphatase, WBC white blood cell count, RBC red blood cell count, SpO2 peripheral oxygen saturation, SBP systolic blood pressure, DBP diastolic blood pressure, RRT renal replacement therapy ● Model 1 was adjusted for age and race ● Model 2 was further adjusted based on Model 1 for severity scores (SOFA), comorbidities (congestive heart failure, peripheral vascular disease, mild liver disease), and laboratory parameters (ALP, WBC, creatinine, RBC, total bilirubin, and sodium) ● Model 3 was further adjusted based on Model 2 for vital signs (SpO2, SBP, DBP, temperature) and treatment modalities (RRT). Model 3 was the primary adjusted model reported in the Results Restricted cubic spline curve of the PNI for 90-day mortality risk. Abbreviations: HR: Hazard ratio; CI: Confidence interval; PNI: Prognostic Nutritional Index Subgroup Analyses To investigate the relationship between PNI low and high PNI groups and 90-day mortality across other categories, we performed further subgroup analyses (Fig. 4). The results revealed a significant interaction between RRT status (Yes/ No) in the correlation between low and high PNI groups and 90-day mortality. Subgroup analyses of the association between low and high PNI groups and 90-day mortality. Abbreviations: SOFA, Sequential Organ Failure Assessment; APS III, Acute Physiology Score III; RRT, Renal Replacement Therapy; HR, hazard ratio; CI, confidence interval. Note: Hazard ratios were estimated using unadjusted Cox proportional hazards models. The figure shows the association between low and high PNI groups and 90-day mortality across different subgroups Secondary outcome The association between the low PNI group and long-term outcomes was consistent with the findings regarding 90-day mortality. The 180-day mortality rate (56.4% vs. 37.1%) and 360-day mortality rate (60.6% vs. 41.5%) in the low PNI group were significantly higher than those in the high PNI group (all p < 0.001; Table 1). Kaplan–Meier survival curves showed that the survival rate in the low PNI group was significantly lower than in the high PNI group at both 180 days and 360 days (log-rank test, p < 0.001, Supplementary Fig. 2). Results from the Cox proportional hazards regression model (Supplementary Table 4) indicate that, even after adjusting for potential confounding variables, the risk of 180-day mortality (adjusted HR: 1.55, 95% CI: 1.35–1.78, p < 0.001) and 360-day mortality (adjusted HR: 1.51, 95% CI: 1.32–1.73, p < 0.001) were significantly elevated. After further reducing confounding using IPTW, the weighted cohort’s Cox regression results showed that the low PNI group was associated with significantly higher risks of 180-day mortality (adjusted HR: 1.48, 95% CI: 1.28–1.71) and 360-day mortality (adjusted HR: 1.44, 95% CI: 1.25–1.65) compared with those in the high-PNI group (both p < 0.001; Supplementary Table 4). RCS analysis confirmed a non-linear relationship between PNI and long-term mortality, and a similar pattern was observed after PNI standardisation (Supplementary Fig. 3 and Supplementary Fig. 5). Both the overall association and the nonlinearity were statistically significant (overall association P < 0.001; nonlinearity P < 0.001). The locations of the nodes and reference points were the same as those used in the analysis of 90-day mortality. ROC curve analyses for 180-day and 360-day mortality yielded findings consistent with those for 90-day mortality, with PNI, albumin, and SOFA demonstrating comparable discriminatory performance (Supplementary Fig. 4). Subgroup analyses showed trends similar to those observed for the primary outcome (Supplementary Figs. 6). Discussion To our knowledge, this is the first study that examines the connection between PNI scores and a poor prognosis in individuals with SAD. Sepsis-associated delirium is a severe condition caused by neuroinflammation and impaired cerebral perfusion, with no specific treatment currently available [41]. It is linked to high death rates and affects up to half of all sepsis patients [42]. Due to the clinical complexity of SAD, identifying accessible biomarkers that reflect systemic vulnerability could provide valuable information for predicting outcomes and complementing existing clinical assessments. The PNI was initially developed to assess nutritional risk in patients undergoing gastrointestinal surgery [43]. However, its prognostic value has been increasingly demonstrated in a number of serious illnesses in recent years. Although PNI was initially developed for surgical risk assessment, it has since demonstrated prognostic significance in several surgical and critical care populations. These include patients with severe ventilator-associated pneumonia, intestinal obstruction, and cardiovascular surgery [23, 24, 44]. Our findings broaden the scope of this application by suggesting that PNI could be used as an additional tool for risk stratification in the context of SAD. To further investigate the relationship between PNI and 90-day mortality, we conducted a RCS analysis. Interestingly, we observed a non-linear association. Below a reference threshold of approximately 30, the risk of death increased sharply as PNI decreased. This suggests that severe nutritional and immune depletion is associated with poorer survival. Conversely, we observed an upward trend in the mortality curve in the higher PNI range (PNI > 30). However, the wide 95% confidence intervals in this range reflect the limited sample size, so this trend should be interpreted with caution. From a physiological perspective, the increased risk associated with high PNI levels does not mean that superior nutritional status is harmful. In fact, an elevated PNI score in the early stages of sepsis likely reflects acute blood concentration or severe hypovolemia. During septic shock, venous vasodilation and capillary leakage lead to relative hypovolemia, and fluid loss causes absolute hypovolemia. Together, these factors result in a decrease in the effective circulating blood volume [45]. Since PNI is calculated based on albumin and lymphocyte counts, acute dehydration or plasma volume contraction can artificially increase these values, leading to a pathologically high PNI score that obscures the patient’s actual compromised condition. This non-linear risk profile is not unique to PNI; it is consistent with findings from other clinical biomarkers. Many biological markers probably have an ideal range, which may be associated with better patient outcomes. Indeed, a recent systematic review of the stress hyperglycaemia ratio (SHR), another widely used metabolic index, revealed a consistent U- or J-shaped association with clinical outcomes in critically ill and septic patients. This demonstrates that both low and high SHR levels are associated with increased mortality [46]. For example, Liu et al. [47]discovered a U-shaped correlation between hematocrit minus albumin (HCT-ALB) and in-hospital mortality in septic patients. They pointed out that elevated HCT-ALB values represent a complex pathological state that includes both the inflammation and malnourishment indicated by hypoalbuminemia and the risk of blood concentration implied by high hematocrit. In clinical practice, PNI may provide additional value for risk stratification and early identification of patients at higher risk of adverse outcomes. The potential pathophysiological mechanisms underlying the association between low PNI and poor outcomes in SAD are likely multifactorial, reflecting a state characterized by nutritional depletion and immune dysregulation. Serum albumin, a key component of PNI, has been shown in clinical studies to be independently associated with mortality in patients with sepsis and to increase susceptibility to infection and severe complications [48]. Mechanistically, excessive secretion of phospholipase A₂ in sepsis catalyzes the hydrolysis of membrane phospholipids. The resulting abundance of fatty acids occupies fatty acid-binding sites on albumin, leading to structural alterations in albumin and loss of its detoxification function. This exacerbates cellular damage [49]. Concurrently, the lymphocyte component of PNI reflects the host’s immune capacity. Significant depletion of T and B cells in sepsis is also closely associated with increased mortality [50]. Importantly, common clinical interventions in the early management of sepsis may interfere with the interpretation of PNI. Current guidelines emphasize aggressive fluid resuscitation during the initial phase of sepsis treatment [51]. This rapid infusion often leads to acute hemodilution, thereby reducing serum albumin concentrations [52]. Simultaneously, sepsis-induced systemic capillary leak syndrome causes albumin to leak into the interstitial space, further exacerbating the loss of intravascular albumin [53]. Therefore, low PNI scores in the early stages may reflect fluid overload and severe systemic capillary leak syndrome rather than preexisting malnutrition. Since positive fluid balance is an independent risk factor for poor prognosis in patients with sepsis [54]. Clinicians should exercise caution when interpreting low PNI scores early on and evaluate them in conjunction with the patient’s fluid resuscitation volume and volume status. Future prospective studies are needed to assess whether monitoring PNI trends after volume stabilization provides superior prognostic value compared to a single baseline measurement. Specifically, early exogenous albumin administration to sepsis patients may elevate the incidence of SA-AKI while providing no short-term survival benefits, according to a recent real-world targeted trial simulation [55]. This underscores that the malnutrition and immunosuppression indicated by low PNI may be better managed through comprehensive nutritional support and immunomodulatory strategies rather than simple albumin supplementation. In fact, two recent reviews support shifting the treatment approach. Stoian et al. emphasize that serum albumin serves as both a nutritional indicator and a marker of acute disease severity. This reinforces the importance of personalized nutrition strategies in the ICU [56]. Concurrently, Oami et al. argue that critical care nutrition should shift toward “metabolic modulation,” which utilizes targeted enteral and microbiota-based therapies to regulate systemic inflammation instead of merely correcting nutrient deficiencies [57]. In general, treatment plans for patients with low PNI should be tailored to their specific circumstances rather than relying solely on protein supplementation. Finally, PNI should not be viewed as a substitute for existing acute severity scores, such as SOFA, which are advantageous in capturing acute organ dysfunction and immediate physiological disturbances. Further, the ROC analysis suggests that PNI provides comparable prognostic information to SOFA and albumin. This supports the use of PNI as a complementary marker rather than a replacement for established clinical assessment tools. PNI can serve as a simple, cost-effective supplemental measure reflecting a patient’s nutritional and immune reserves. This adds a useful dimension to the overall assessment of patients with SAD. Additionally, subgroup analysis indicates that the relationship between PNI and mortality could differ among patients receiving RRT. Notably, the prognostic effect of PNI appears attenuated in patients receiving RRT. One possible explanation is that RRT is associated with changes in the physiological interpretation of PNI through mechanisms such as fluid shifts, extracorporeal losses, and immunological and inflammatory changes. These mechanisms are associated with dynamic changes in albumin and lymphocyte levels. Moreover, patients requiring RRT tend to have more severe systemic disease and pronounced multi-organ dysfunction, which may make it more difficult to assess the association between nutritional and immune status and prognosis. This study has several limitations. First, due to the retrospective observational design, causality between PNI and outcomes in patients with SAD cannot be established. Second, this was a single-center study based on the MIMIC-IV database, which may limit the generalizability of the findings to other healthcare systems and populations. Third, PNI was assessed only using early ICU measurements, and longitudinal changes in nutritional–immune status during hospitalization were not evaluated. In addition, serum albumin levels may be influenced by systemic inflammation, fluid resuscitation, or albumin administration, which could affect the accuracy of PNI as a nutritional marker despite the use of early laboratory measurements and IPTW adjustment to mitigate related confounding. Due to database limitations, we were unable to adjust for 24-hour fluid balance directly. Fourth, although extensive multivariable adjustment was performed, residual confounding factors, including nutritional interventions and delirium management strategies, cannot be completely excluded. Fifth, patients with missing albumin or lymphocyte measurements were excluded from the analysis, which may have introduced some degree of selection bias. Finally, delirium subtypes could not be differentiated because subtype-specific assessments were unavailable in the database. In particular, hypoactive delirium may have been underrecognized in routine clinical documentation, which could have affected patient classification and outcome interpretation. Prospective multicenter studies are needed to further validate and extend these findings. Conclusions Among patients with sepsis-associated delirium in the MIMIC-IV database, a lower PNI was independently associated with higher 90-day mortality. Nonlinear modeling suggested a possible nonlinear association, although this finding should be interpreted cautiously. As an inexpensive and readily calculated marker derived from routine laboratory parameters, PNI is a simple, useful, and easily accessible tool that can help identify patients with SAD who are at increased risk of adverse outcomes at an early stage. However, because our findings are based solely on a single-center database, further validation in independent, multi-center ICU cohorts is required to confirm these results. Supplementary Information Supplementary Material 1. Supplementary Table 1 Baseline characteristics by 90-day survival status and Prognostic Nutritional Index (PNI). Supplementary Table 2 Baseline characteristics of included and excluded patients in the study cohort. Supplementary Table 3 Baseline Characteristics of the Study Population After Inverse Probability of Treatment Weighting. Supplementary Table 4 Cox regression analysis of the association between PNI and 180-day and 360-day mortality in sepsis-associated delirium patients. Supplementary Figure 1 Standardized Mean Differences Before and After Inverse Probability of Treatment Weighting. Supplementary Figure 2 Kaplan-Meier curves for mortality by PNI groups. (A), 180-day (B), and 360-day. Supplementary Figure 3 Restricted cubic spline curves for standardized PNI and mortality risk. (A), 90-day (B), 180-day (C), and 360-day. Supplementary Figure 4 ROC Curves for PNI in Predicting 90‑day, 180‑day, and 360‑day All‑cause Mortality. Supplementary Figure 5 Restricted cubic spline curves for PNI and mortality risk. (A), 180-day (B), and 360-day. Supplementary Figure 6 Subgroup analyses of the association between low and high PNI groups and mortality. (A), 180-day (B), and 360-day. Acknowledgements The authors would like to thank the team of the Laboratory for Computational Physiology from the Massachusetts Institute of Technology for the development and maintenance of the MIMIC-IV databases.This work is supported by the Critical care medical data subsystem of Kunlun Big Data Management System technically. Authors’ contributions Shengnan Kong: Writing – original draft, Formal analysis, Conceptualization; Chunling Yang: Writing – original draft, Formal analysis; Xinya Li: Data curation, Validation; Qin Lai: Investigation. Jiafang Zhang: Investigation; Yu Wang: Writing – review & editing, Supervision, Conceptualization; Jun Lyu: Writing – review & editing, Supervision, Conceptualization. Funding This work was supported by the Science and Technology Projects in Guangzhou (2025A03J4246). Data availability The data were available on the MIMIC-IV website at https://mimic.physionet.org/. The data in this article can be reasonably applied to the corresponding author. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Data extracted from the MIMIC-IV database do not require individual informed consent because the research data is publicly available, and all patient data are de-identified. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Shengnan Kong and Chunling Yang contributed equally to this work. Contributor Information Yu Wang, Email: hulibuwangyu@126.com. Jun Lyu, Email: lyujun2020@jnu.edu.cn. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Table 1 Baseline characteristics by 90-day survival status and Prognostic Nutritional Index (PNI). Supplementary Table 2 Baseline characteristics of included and excluded patients in the study cohort. Supplementary Table 3 Baseline Characteristics of the Study Population After Inverse Probability of Treatment Weighting. Supplementary Table 4 Cox regression analysis of the association between PNI and 180-day and 360-day mortality in sepsis-associated delirium patients. Supplementary Figure 1 Standardized Mean Differences Before and After Inverse Probability of Treatment Weighting. Supplementary Figure 2 Kaplan-Meier curves for mortality by PNI groups. (A), 180-day (B), and 360-day. Supplementary Figure 3 Restricted cubic spline curves for standardized PNI and mortality risk. (A), 90-day (B), 180-day (C), and 360-day. Supplementary Figure 4 ROC Curves for PNI in Predicting 90‑day, 180‑day, and 360‑day All‑cause Mortality. Supplementary Figure 5 Restricted cubic spline curves for PNI and mortality risk. (A), 180-day (B), and 360-day. Supplementary Figure 6 Subgroup analyses of the association between low and high PNI groups and mortality. (A), 180-day (B), and 360-day. Data Availability Statement The data were available on the MIMIC-IV website at https://mimic.physionet.org/. The data in this article can be reasonably applied to the corresponding author.