Dynamic prognostic nutritional index and 28-day mortality in sepsis: a joint modeling analysis Authors: Feng Xu, Jinjiang Li, Wei Zhao, Shijing Ding, Jing Cao Published: 2026 ABSTRACT BackgroundAlthough the Prognostic Nutritional Index (PNI) is associated with sepsis outcomes, admission values may primarily reflect acute illness severity. We examined whether longitudinal PNI provides prognostic value beyond baseline risk.MethodsIn this retrospective cohort study of 492 adults with sepsis, baseline PNI was assessed using Kaplan–Meier analysis and adjusted restricted cubic splines. A joint model was used to link repeated PNI measurements from days 1 to 7 with time to 28-day mortality while accounting for informative dropout due to death or discharge. Time-dependent Cox models were further used to compare dynamic PNI with albumin and total lymphocyte count.ResultsBaseline PNI showed a non-significant trend toward survival separation (log-rank p = 0.063) and was not independently associated with 28-day mortality after multivariable adjustment (p = 0.122). By contrast, dynamic PNI was independently associated with mortality in time-dependent analyses (HR 0.730, p = 0.032), whereas albumin and total lymphocyte count were not. Joint modeling likewise demonstrated a significant independent association between longitudinal PNI and mortality (HR 0.831, p < 0.001).ConclusionBaseline PNI showed limited independent prognostic value, whereas longitudinal PNI provided additional prognostic information and showed modestly better predictive performance than its individual components.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12986-026-01137-w. Introduction Sepsis is defined as life-threatening organ dysfunction caused by a dysregulated host response to infection. It is characterized by profound physiological disturbances, including hypercatabolism, systemic inflammation, and immune suppression [1]. These processes are closely associated with adverse clinical outcomes, including prolonged hospitalization, increased susceptibility to secondary infections, and elevated mortality [2]. The Prognostic Nutritional Index (PNI), calculated from serum albumin concentration and total lymphocyte count, was originally developed to stratify perioperative risk [3]. In critically ill patients, however, both components are strongly influenced by acute inflammation and organ dysfunction rather than nutritional status alone [4]. In sepsis, hypoalbuminemia largely reflects capillary leakage, fluid shifts, and altered hepatic synthesis, whereas lymphopenia is a hallmark of sepsis-associated immune dysfunction and lymphocyte apoptosis [5]. Therefore, in the ICU setting, PNI should be interpreted as an accessible composite marker reflecting inflammation-related protein kinetics and immune status, rather than solely as an indicator of malnutrition [6]. Previous studies have reported associations between low admission PNI and mortality in sepsis [7]. However, reliance on a single baseline measurement has important limitations [8]. Admission albumin and lymphocyte values are strongly influenced by initial disease severity and early resuscitation, and may therefore provide limited incremental information beyond established severity scores such as APACHE II and SOFA [9]. In addition, sepsis is a dynamic syndrome, and patients with similar baseline profiles may subsequently follow markedly different clinical trajectories, ranging from rapid recovery to persistent immune and metabolic dysfunction [10]. Longitudinal biomarker patterns may better capture disease evolution and treatment response than isolated measurements [11]. Although recent studies have examined dynamic biomarker changes in critical care, most have relied on conventional methods such as repeated-measures ANOVA or simple time-dependent Cox models [12, 13]. These approaches may be limited because they do not fully account for informative dropout due to death or discharge [14]. In severe sepsis, patients who die early inevitably contribute fewer longitudinal observations, which may bias estimation of biomarker trajectories [15]. Joint modeling provides a principled framework for linking repeated biomarker measurements with time-to-event outcomes and can help mitigate this problem [16]. Although joint modeling has been increasingly applied in oncology and chronic disease research, its use in acute critical care remains limited [17], particularly for evaluating composite nutritional and immune-related markers such as PNI in sepsis. Therefore, in this retrospective cohort of adult patients with sepsis, we aimed to address these methodological and clinical gaps. Specifically, we applied a joint modeling approach to evaluate longitudinal PNI trajectories while accounting for informative dropout related to death or discharge. We further examined whether dynamic PNI provided prognostic information beyond baseline illness severity and compared its prognostic performance with that of its individual components. Finally, we explored whether the prognostic information conveyed by dynamic PNI was more closely related to its absolute level over time or to change from baseline. Method Study design and participants We conducted a single-center retrospective cohort study to evaluate the associations of baseline and longitudinal Prognostic Nutritional Index (PNI) with 28-day mortality in adult patients with sepsis. Consecutive eligible patients admitted to the intensive care unit (ICU) of the First Hospital of Shanxi Medical University between January 2021 and December 2025 were screened for inclusion. The inclusion criteria were as follows: (1) age ≥ 18 years; (2) diagnosis of sepsis according to the Sepsis-3 criteria; and (3) availability of at least one paired measurement of serum albumin and total lymphocyte count within 24 h of ICU admission for calculation of baseline PNI. Longitudinal PNI values were derived from routine laboratory tests obtained during the ICU stay. In our ICU, serum albumin and lymphocyte counts are typically measured daily as part of standard care. Accordingly, PNI was calculated once daily from Day 1 (ICU admission) through Day 7, or until death or discharge, whichever occurred first. The exclusion criteria were: (1) pregnancy or lactation; (2) profound pre-existing immunosuppression, including hematologic malignancy, solid-organ transplantation, HIV infection, long-term systemic immunosuppressive therapy, or recent chemotherapy; (3) receipt of sustained high-dose albumin supplementation during the observation period, predefined as > 40 g/day for at least 2 consecutive days as an operational criterion, because such supplementation was considered likely to materially alter serial serum albumin levels and thus complicate interpretation of longitudinal PNI trajectories; and (4) incomplete data precluding outcome ascertainment or absence of paired serum albumin and total lymphocyte count measurements within the first 24 h of ICU admission. During the study period, 557 eligible patients with sepsis were identified. Among them, 65 were excluded because baseline PNI components were not available within the predefined 24-hour window, leaving a final analytical cohort of 492 patients. The study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of the First Hospital of Shanxi Medical University (Approval No. NO.KYLL-2026-040). The requirement for informed consent was waived because of the retrospective observational design. Data collection Clinical data were retrospectively extracted from the electronic medical record system. Baseline variables included demographic characteristics (age and sex), comorbidities (hypertension, coronary artery disease, diabetes mellitus, chronic obstructive pulmonary disease, chronic kidney disease, and malignancy), and illness severity indices at ICU admission, including the Acute Physiology and Chronic Health Evaluation II (APACHE II) score and the Sequential Organ Failure Assessment (SOFA) score. Use of mechanical ventilation and renal replacement therapy during the ICU stay was also recorded. Definition and measurement of PNI The Prognostic Nutritional Index (PNI) was calculated using the standard formula: PNI = serum albumin (g/L) + 5 × total lymphocyte count (10^9/L). Baseline PNI was defined using the first available paired measurements of serum albumin and total lymphocyte count obtained within 24 h of ICU admission (Day 1). Longitudinal PNI values were calculated from routine laboratory tests performed during the ICU stay. Outcome definition The primary endpoint was 28-day all-cause mortality. We selected 28-day mortality because it is a commonly used short-term outcome in sepsis research and is clinically aligned with the acute disease course and the first-week longitudinal biomarker assessment in the present study. Survival time was measured from ICU admission (Day 1) to the date of death, with survivors censored at day 28. Statistical analysis Continuous variables were summarized as mean (standard deviation) or median (interquartile range), as appropriate, and categorical variables were presented as counts (percentages). All analyses focused on 28-day mortality. Baseline PNI was categorized into quartiles and compared using Kaplan–Meier survival curves with the log-rank test. To further assess the functional form of the association between baseline PNI and mortality, we fitted an adjusted restricted cubic spline (RCS) model and tested for non-linearity using the Wald test. To evaluate the prognostic value of longitudinal PNI while accounting for informative dropout due to death or discharge, we fitted a joint model linking longitudinal and survival data. The longitudinal submodel consisted of a linear mixed-effects model for repeated PNI measurements from days 1 to 7, including fixed effects for time, age, sex, APACHE II score, and baseline SOFA score, together with subject-specific random intercepts and slopes. The survival submodel used a Weibull proportional hazards formulation adjusted for age, sex, baseline severity, organ support, and comorbidities. A Weibull model was selected because it provides a flexible parametric specification for the baseline hazard and is commonly used in joint modeling of longitudinal and survival data. The longitudinal and survival processes were primarily linked using a time-dependent current-value parameterization. To further assess whether the prognostic information conveyed by dynamic PNI was more closely related to its absolute level over time or to change from baseline, we additionally fitted an alternative change-from-baseline parameterization. In this model, both baseline PNI and the subject-specific longitudinal change in PNI (ΔPNI) were entered separately into the survival submodel. For interpretability, association parameters were scaled to report hazard ratios (HRs) and 95% confidence intervals (CIs) per 10-point increase in current PNI or ΔPNI. In addition, time-dependent Cox regression models were fitted to compare dynamic PNI with albumin and total lymphocyte count individually. In these models, SOFA score was treated as a time-varying covariate. Model discrimination and relative fit were assessed using Harrell’s C-index and the Akaike Information Criterion (AIC), respectively. All tests were two-sided, and a p value < 0.05 was considered statistically significant. All analyses were performed in R software using the survival, nlme, and JM packages. Result Baseline characteristics During the study period, 557 eligible adult patients with sepsis were identified. Among them, 65 patients (11.7%) were excluded from the primary analysis because paired baseline PNI components were not available within the first 24 h of ICU admission (Supplementary Table S1). To explore the potential for selection bias, we compared the baseline characteristics of the included cohort (n = 492) and the excluded cohort (n = 65). No statistically significant differences were observed in age, sex, illness severity (APACHE II, p = 0.119; SOFA, p = 0.252), life-support requirements, or major comorbidities (all p > 0.05). The final analytical cohort comprised 492 patients with sepsis, including 273 (55.5%) 28-day survivors and 219 (44.5%) non-survivors. Baseline characteristics are summarized in Table 1. Compared with survivors, non-survivors were older (median 70.0 vs. 63.0 years, p < 0.001) and had higher illness severity at ICU admission, as reflected by APACHE II scores (24.0 vs. 23.0, p < 0.001) and SOFA scores (9.0 vs. 7.0, p < 0.001). Non-survivors also had lower baseline PNI values than survivors (32.55 [IQR 28.60–35.00] vs. 32.90 [IQR 30.05–35.90], p = 0.018). In addition, non-survivors had a higher prevalence of chronic obstructive pulmonary disease (12.8% vs. 6.2%, p = 0.019) and more frequently required organ support, including renal replacement therapy (49.8% vs. 28.6%, p < 0.001) and mechanical ventilation (85.8% vs. 57.9%, p < 0.001). Table 1Baseline characteristics of the study population according to 28-day survival statusCharacteristicOverallSurvivorNon-SurvivorP valuen492273219Age66.0 [57.0, 74.0]63.0 [54.0, 71.0]70.0 [60.0, 76.50]< 0.001Sex = Male (%)278 (56.5)143 (52.4)135 (61.6)0.049APACHE II24.0 [18.0, 30.0]23.0 [16.0, 28.0]24.0 [21.5, 31.0]< 0.001SOFA8.0 [5.0, 11.0]7.0 [4.0, 10.0]9.0 [6.0, 12.0]< 0.001PNI32.55 [29.31, 35.60]32.90 [30.05, 35.90]32.55 [28.60, 35.00]0.018Hypertension = Yes (%)291 (59.1)152 (55.7)139 (63.5)0.081Diabetes = Yes (%)156 (31.7)84 (30.8)72 (32.9)0.618CAD = Yes (%)98 (19.9)47 (17.2)51 (23.3)0.118COPD = Yes (%)45 (9.1)17 (6.2)28 (12.8)0.019CKD = Yes (%)198 (40.2)100 (36.6)98 (44.7)0.083Malignancy = Yes (%)157 (31.9)83 (30.4)74 (33.8)0.482RRT = Yes (%)187 (38.0)78 (28.6)109 (49.8)< 0.001Mech Vent = Yes (%)346 (70.3)158 (57.9)188 (85.8)< 0.001Data are presented as n (%) or median [interquartile range]. Abbreviations: APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; PNI, Prognostic Nutritional Index; CAD, coronary artery disease; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease; RRT, renal replacement therapy; Mech Vent, mechanical ventilation Table 1Baseline characteristics of the study population according to 28-day survival statusCharacteristicOverallSurvivorNon-SurvivorP valuen492273219Age66.0 [57.0, 74.0]63.0 [54.0, 71.0]70.0 [60.0, 76.50]< 0.001Sex = Male (%)278 (56.5)143 (52.4)135 (61.6)0.049APACHE II24.0 [18.0, 30.0]23.0 [16.0, 28.0]24.0 [21.5, 31.0]< 0.001SOFA8.0 [5.0, 11.0]7.0 [4.0, 10.0]9.0 [6.0, 12.0]< 0.001PNI32.55 [29.31, 35.60]32.90 [30.05, 35.90]32.55 [28.60, 35.00]0.018Hypertension = Yes (%)291 (59.1)152 (55.7)139 (63.5)0.081Diabetes = Yes (%)156 (31.7)84 (30.8)72 (32.9)0.618CAD = Yes (%)98 (19.9)47 (17.2)51 (23.3)0.118COPD = Yes (%)45 (9.1)17 (6.2)28 (12.8)0.019CKD = Yes (%)198 (40.2)100 (36.6)98 (44.7)0.083Malignancy = Yes (%)157 (31.9)83 (30.4)74 (33.8)0.482RRT = Yes (%)187 (38.0)78 (28.6)109 (49.8)< 0.001Mech Vent = Yes (%)346 (70.3)158 (57.9)188 (85.8)< 0.001Data are presented as n (%) or median [interquartile range]. Abbreviations: APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; PNI, Prognostic Nutritional Index; CAD, coronary artery disease; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease; RRT, renal replacement therapy; Mech Vent, mechanical ventilation Baseline characteristics of the study population according to 28-day survival status Data are presented as n (%) or median [interquartile range]. Abbreviations: APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; PNI, Prognostic Nutritional Index; CAD, coronary artery disease; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease; RRT, renal replacement therapy; Mech Vent, mechanical ventilation Association between baseline PNI and 28-day mortality We first evaluated the association between baseline PNI measured at ICU admission and 28-day mortality. Kaplan–Meier analysis showed a trend toward lower survival probabilities across lower baseline PNI quartiles, although the difference did not reach statistical significance (log-rank p = 0.063). Consistently, adjusted restricted cubic spline (RCS) analysis suggested an approximately linear inverse association between baseline PNI and mortality risk, with no evidence of non-linearity (p for non-linearity = 0.357). However, the 95% confidence intervals crossed the reference line across the observed range, indicating that this association was not statistically significant (Fig. 1). Fig. 1Association between baseline Prognostic Nutritional Index and 28-day mortality. (A) Kaplan–Meier survival curves stratified by baseline PNI quartiles. Differences among quartiles were not statistically significant (log-rank p = 0.063). (B) Adjusted restricted cubic spline analysis of the association between continuous baseline PNI and the hazard ratio for 28-day mortality. The model was adjusted for age, sex, APACHE II score, SOFA score, organ support, and comorbidities. The solid red line indicates the estimated hazard ratio, the blue shaded area indicates the 95% confidence interval, and the dashed grey line indicates the reference hazard ratio of 1.0 (p for non-linearity = 0.357) Association between baseline Prognostic Nutritional Index and 28-day mortality. (A) Kaplan–Meier survival curves stratified by baseline PNI quartiles. Differences among quartiles were not statistically significant (log-rank p = 0.063). (B) Adjusted restricted cubic spline analysis of the association between continuous baseline PNI and the hazard ratio for 28-day mortality. The model was adjusted for age, sex, APACHE II score, SOFA score, organ support, and comorbidities. The solid red line indicates the estimated hazard ratio, the blue shaded area indicates the 95% confidence interval, and the dashed grey line indicates the reference hazard ratio of 1.0 (p for non-linearity = 0.357) In univariable Cox proportional hazards analysis (Table 2), baseline PNI was significantly associated with 28-day mortality (HR 0.744; 95% CI 0.574–0.970; p = 0.031). However, after adjustment for age, sex, APACHE II score, SOFA score, organ support, and comorbidities, baseline PNI was no longer independently associated with 28-day mortality (Model 3: HR 0.808; 95% CI 0.624–1.061; p = 0.122). Table 2Association between baseline Prognostic Nutritional Index and 28-day all-cause mortality in Cox proportional hazards modelsModelHR (95% CI)P valueAdjustmentsCrude Model0.744 (0.574–0.970)0.031NoneModel 10.793 (0.612–1.041)0.091Age, SexModel 20.808 (0.624–1.051)0.113Model 1 + APACHE II, Baseline SOFA, RRT, Mech VentModel 30.808 (0.624–1.061)0.122Model 2 + Comorbidities (HTN, DM, CAD, COPD, CKD, Malignancy)Hazard ratios (HRs) are reported per 10-point increase in PNI. Abbreviations: HR, hazard ratio; CI, confidence interval; PNI, Prognostic Nutritional Index; APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; RRT, renal replacement therapy; Mech Vent, mechanical ventilation; HTN, hypertension; DM, diabetes mellitus; CAD, coronary artery disease; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease Table 2Association between baseline Prognostic Nutritional Index and 28-day all-cause mortality in Cox proportional hazards modelsModelHR (95% CI)P valueAdjustmentsCrude Model0.744 (0.574–0.970)0.031NoneModel 10.793 (0.612–1.041)0.091Age, SexModel 20.808 (0.624–1.051)0.113Model 1 + APACHE II, Baseline SOFA, RRT, Mech VentModel 30.808 (0.624–1.061)0.122Model 2 + Comorbidities (HTN, DM, CAD, COPD, CKD, Malignancy)Hazard ratios (HRs) are reported per 10-point increase in PNI. Abbreviations: HR, hazard ratio; CI, confidence interval; PNI, Prognostic Nutritional Index; APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; RRT, renal replacement therapy; Mech Vent, mechanical ventilation; HTN, hypertension; DM, diabetes mellitus; CAD, coronary artery disease; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease Association between baseline Prognostic Nutritional Index and 28-day all-cause mortality in Cox proportional hazards models Hazard ratios (HRs) are reported per 10-point increase in PNI. Abbreviations: HR, hazard ratio; CI, confidence interval; PNI, Prognostic Nutritional Index; APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; RRT, renal replacement therapy; Mech Vent, mechanical ventilation; HTN, hypertension; DM, diabetes mellitus; CAD, coronary artery disease; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease Dynamic PNI and time-dependent analysis Given the dynamic nature of sepsis, we further evaluated the association between longitudinal PNI and 28-day mortality. Assessment of the proportional hazards assumption did not indicate a violation for either the overall model (p = 0.153) or the time-dependent PNI covariate (p = 0.774). In contrast to the baseline analysis, time-dependent PNI remained independently associated with mortality after multivariable adjustment (Table 3). In the fully adjusted model (Model 3), each 10-point increase in time-dependent PNI was associated with a 27% lower hazard of 28-day mortality (HR 0.730; 95% CI 0.550–0.970; p = 0.032). Table 3Association between dynamic Prognostic Nutritional Index and 28-day mortality in time-dependent Cox proportional hazards modelsModelHR (95% CI)P valueAdjustmentsCrude Model0.651 (0.489–0.860)0.002NoneModel 10.722 (0.539–0.961)0.025Age, SexModel 20.730 (0.550–0.970)0.028Model 1 + APACHE II, Time-varying SOFA, RRT, Mech VentModel 30.730 (0.550–0.970)0.032Model 2 + Comorbidities (HTN, DM, CAD, COPD, CKD, Malignancy) Hazard ratios (HRs) are reported per 10-point increase in PNI. Abbreviations: HR, hazard ratio; CI, confidence interval; PNI, Prognostic Nutritional Index; APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; RRT, renal replacement therapy; Mech Vent, mechanical ventilation; HTN, hypertension; DM, diabetes mellitus; CAD, coronary artery disease; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease Table 3Association between dynamic Prognostic Nutritional Index and 28-day mortality in time-dependent Cox proportional hazards modelsModelHR (95% CI)P valueAdjustmentsCrude Model0.651 (0.489–0.860)0.002NoneModel 10.722 (0.539–0.961)0.025Age, SexModel 20.730 (0.550–0.970)0.028Model 1 + APACHE II, Time-varying SOFA, RRT, Mech VentModel 30.730 (0.550–0.970)0.032Model 2 + Comorbidities (HTN, DM, CAD, COPD, CKD, Malignancy) Hazard ratios (HRs) are reported per 10-point increase in PNI. Abbreviations: HR, hazard ratio; CI, confidence interval; PNI, Prognostic Nutritional Index; APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; RRT, renal replacement therapy; Mech Vent, mechanical ventilation; HTN, hypertension; DM, diabetes mellitus; CAD, coronary artery disease; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease Association between dynamic Prognostic Nutritional Index and 28-day mortality in time-dependent Cox proportional hazards models Hazard ratios (HRs) are reported per 10-point increase in PNI. Abbreviations: HR, hazard ratio; CI, confidence interval; PNI, Prognostic Nutritional Index; APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; RRT, renal replacement therapy; Mech Vent, mechanical ventilation; HTN, hypertension; DM, diabetes mellitus; CAD, coronary artery disease; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease We further compared the performance of dynamic PNI with that of its individual components (Table 4). In multivariable time-dependent Cox analyses, dynamic PNI remained significantly associated with mortality (p = 0.032), whereas dynamic albumin (p = 0.065) and total lymphocyte count (p = 0.537) were not. In addition, the time-dependent PNI model showed numerically the highest C-index (0.676) and the lowest AIC (2414.8) among the three time-dependent models. However, these differences in model performance were modest. Table 4Comparison of model performance for Prognostic Nutritional Index and its individual components in baseline and time-dependent Cox regression modelsModel TypeMarkerHR (95% CI)P valueC-index (95% CI)AICBaseline CoxAlbumin0.971 (0.942–1.001)0.0590.666 (0.630–0.701)2543.7Lymphocytes1.012 (0.770–1.329)0.9340.661 (0.625–0.697)2547.3PNI0.808 (0.625–1.061)0.1220.664 (0.628–0.699)2544.9Time-Dependent CoxAlbumin0.968 (0.935–1.002)0.0650.675 (0.639–0.711)2416.0Lymphocytes0.924 (0.718–1.188)0.5370.670 (0.635–0.706)2419.0PNI0.730 (0.550–0.970)0.0320.676 (0.640–0.712)2414.8All models were adjusted for age, sex, APACHE II score, SOFA score (baseline or time-varying, as appropriate), renal replacement therapy, mechanical ventilation, and comorbidities. Hazard ratios (HRs) for PNI are reported per 10-point increase. HRs for albumin and total lymphocyte count are reported per 1 g/L and 1 × 10^9/L increase, respectively. Abbreviations: HR, hazard ratio; CI, confidence interval; AIC, Akaike Information Criterion; C-index, concordance index; PNI, Prognostic Nutritional Index; APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; RRT, renal replacement therapy Table 4Comparison of model performance for Prognostic Nutritional Index and its individual components in baseline and time-dependent Cox regression modelsModel TypeMarkerHR (95% CI)P valueC-index (95% CI)AICBaseline CoxAlbumin0.971 (0.942–1.001)0.0590.666 (0.630–0.701)2543.7Lymphocytes1.012 (0.770–1.329)0.9340.661 (0.625–0.697)2547.3PNI0.808 (0.625–1.061)0.1220.664 (0.628–0.699)2544.9Time-Dependent CoxAlbumin0.968 (0.935–1.002)0.0650.675 (0.639–0.711)2416.0Lymphocytes0.924 (0.718–1.188)0.5370.670 (0.635–0.706)2419.0PNI0.730 (0.550–0.970)0.0320.676 (0.640–0.712)2414.8All models were adjusted for age, sex, APACHE II score, SOFA score (baseline or time-varying, as appropriate), renal replacement therapy, mechanical ventilation, and comorbidities. Hazard ratios (HRs) for PNI are reported per 10-point increase. HRs for albumin and total lymphocyte count are reported per 1 g/L and 1 × 10^9/L increase, respectively. Abbreviations: HR, hazard ratio; CI, confidence interval; AIC, Akaike Information Criterion; C-index, concordance index; PNI, Prognostic Nutritional Index; APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; RRT, renal replacement therapy Comparison of model performance for Prognostic Nutritional Index and its individual components in baseline and time-dependent Cox regression models All models were adjusted for age, sex, APACHE II score, SOFA score (baseline or time-varying, as appropriate), renal replacement therapy, mechanical ventilation, and comorbidities. Hazard ratios (HRs) for PNI are reported per 10-point increase. HRs for albumin and total lymphocyte count are reported per 1 g/L and 1 × 10^9/L increase, respectively. Abbreviations: HR, hazard ratio; CI, confidence interval; AIC, Akaike Information Criterion; C-index, concordance index; PNI, Prognostic Nutritional Index; APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; RRT, renal replacement therapy Joint modeling of longitudinal PNI and survival To account for informative dropout related to death or discharge during follow-up, we fitted a joint model linking longitudinal PNI measurements and survival data (Table 5). In the longitudinal submodel, PNI trajectories were significantly associated with time (day coefficient 0.456; p < 0.001), as well as with baseline age, sex, and SOFA score. In the primary survival submodel using a current-value parameterization, the current PNI value was significantly associated with 28-day mortality after adjustment for baseline severity and comorbidities (association parameter: -0.0185; p < 0.001). Expressed on a clinically interpretable scale, each 10-point increase in current PNI was associated with a lower hazard of 28-day mortality (HR 0.831; 95% CI 0.755–0.915; p < 0.001). To further explore whether the prognostic information conveyed by dynamic PNI was more closely related to its absolute level over time or to change from baseline, we additionally fitted an alternative change-from-baseline parameterization (Table 5). In this model, both baseline PNI and the subsequent change in PNI over time (ΔPNI) were independently associated with 28-day mortality. Specifically, each 10-point increase in ΔPNI from baseline was associated with a lower mortality hazard (HR 0.882; 95% CI 0.822–0.946; p < 0.001). Table 5Joint modeling of longitudinal PNI trajectories and 28-day mortality using current-value and change-from-baseline parameterizationsVariableEstimate (β)Std. ErrorP valueHR (95% CI) A. Longitudinal Submodel a (Intercept)40.6060.433< 0.001—Day (Time slope)0.4560.044< 0.001—Age-0.0790.006< 0.001—Sex (Male)-1.4520.181< 0.001—APACHE II-0.0190.0120.124—Baseline SOFA-0.130.027< 0.001— B. Survival Submodel b Age0.0130.0050.0111.013 (1.003–1.024)Sex (Male)0.1110.1480.4541.118 (0.836–1.495)APACHE II0.0060.0090.5171.006 (0.988–1.025)Baseline SOFA0.0210.0330.5291.021 (0.957–1.089)RRT0.420.210.0451.522 (1.009–2.295)Mechanical ventilation0.9850.23< 0.0012.679 (1.706–4.207)Hypertension0.4840.3240.1361.622 (0.859–3.060)Diabetes-0.2470.2570.3350.781 (0.472–1.291)CAD0.2070.1710.2251.230 (0.880–1.721)COPD0.3160.2270.1641.371 (0.879–2.137)CKD0.140.1460.3371.151 (0.864–1.532)Malignancy-0.1010.1560.5180.904 (0.666–1.227) C. Association Parameter c Primary Model: Current Value ParameterizationCurrent PNI Value-0.0190.005< 0.0010.982 (0.972–0.991)Per 10-point higher PNI——< 0.0010.831 (0.755–0.915)Alternative Model: Change-from-Baseline ParameterizationBaseline PNI (Day 1)-0.0510.0150.0010.600 (0.448–0.803)ΔPNI-0.0130.004< 0.0010.882 (0.822–0.946) D. Model Parameters Weibull log(shape)-0.6450.075< 0.001—(a) The longitudinal submodel (linear mixed-effects model) estimates the evolution of PNI over time. (b) The survival submodel (Weibull relative-risk model) estimates time to death, adjusted for the listed baseline clinical covariates. (c) In the primary model, the association parameter represents the effect of the instantaneous current PNI value on mortality risk. For interpretability, the corresponding hazard ratio for a 10-point higher current PNI is additionally shown as exp(association × 10). In the alternative model, baseline PNI and ΔPNI are entered separately into the survival submodel; hazard ratios for these variables are reported per 10-point increase. Abbreviations: PNI, Prognostic Nutritional Index; ΔPNI, change in PNI from baseline; HR, hazard ratio; CI, confidence interval; APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; RRT, renal replacement therapy; CAD, coronary artery disease; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease Table 5Joint modeling of longitudinal PNI trajectories and 28-day mortality using current-value and change-from-baseline parameterizationsVariableEstimate (β)Std. ErrorP valueHR (95% CI) A. Longitudinal Submodel a (Intercept)40.6060.433< 0.001—Day (Time slope)0.4560.044< 0.001—Age-0.0790.006< 0.001—Sex (Male)-1.4520.181< 0.001—APACHE II-0.0190.0120.124—Baseline SOFA-0.130.027< 0.001— B. Survival Submodel b Age0.0130.0050.0111.013 (1.003–1.024)Sex (Male)0.1110.1480.4541.118 (0.836–1.495)APACHE II0.0060.0090.5171.006 (0.988–1.025)Baseline SOFA0.0210.0330.5291.021 (0.957–1.089)RRT0.420.210.0451.522 (1.009–2.295)Mechanical ventilation0.9850.23< 0.0012.679 (1.706–4.207)Hypertension0.4840.3240.1361.622 (0.859–3.060)Diabetes-0.2470.2570.3350.781 (0.472–1.291)CAD0.2070.1710.2251.230 (0.880–1.721)COPD0.3160.2270.1641.371 (0.879–2.137)CKD0.140.1460.3371.151 (0.864–1.532)Malignancy-0.1010.1560.5180.904 (0.666–1.227) C. Association Parameter c Primary Model: Current Value ParameterizationCurrent PNI Value-0.0190.005< 0.0010.982 (0.972–0.991)Per 10-point higher PNI——< 0.0010.831 (0.755–0.915)Alternative Model: Change-from-Baseline ParameterizationBaseline PNI (Day 1)-0.0510.0150.0010.600 (0.448–0.803)ΔPNI-0.0130.004< 0.0010.882 (0.822–0.946) D. Model Parameters Weibull log(shape)-0.6450.075< 0.001—(a) The longitudinal submodel (linear mixed-effects model) estimates the evolution of PNI over time. (b) The survival submodel (Weibull relative-risk model) estimates time to death, adjusted for the listed baseline clinical covariates. (c) In the primary model, the association parameter represents the effect of the instantaneous current PNI value on mortality risk. For interpretability, the corresponding hazard ratio for a 10-point higher current PNI is additionally shown as exp(association × 10). In the alternative model, baseline PNI and ΔPNI are entered separately into the survival submodel; hazard ratios for these variables are reported per 10-point increase. Abbreviations: PNI, Prognostic Nutritional Index; ΔPNI, change in PNI from baseline; HR, hazard ratio; CI, confidence interval; APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; RRT, renal replacement therapy; CAD, coronary artery disease; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease Joint modeling of longitudinal PNI trajectories and 28-day mortality using current-value and change-from-baseline parameterizations (a) The longitudinal submodel (linear mixed-effects model) estimates the evolution of PNI over time. (b) The survival submodel (Weibull relative-risk model) estimates time to death, adjusted for the listed baseline clinical covariates. (c) In the primary model, the association parameter represents the effect of the instantaneous current PNI value on mortality risk. For interpretability, the corresponding hazard ratio for a 10-point higher current PNI is additionally shown as exp(association × 10). In the alternative model, baseline PNI and ΔPNI are entered separately into the survival submodel; hazard ratios for these variables are reported per 10-point increase. Abbreviations: PNI, Prognostic Nutritional Index; ΔPNI, change in PNI from baseline; HR, hazard ratio; CI, confidence interval; APACHE II, Acute Physiology and Chronic Health Evaluation II; SOFA, Sequential Organ Failure Assessment; RRT, renal replacement therapy; CAD, coronary artery disease; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease Discussion This study provides a comprehensive evaluation of the Prognostic Nutritional Index (PNI) in sepsis and suggests a difference in prognostic relevance between admission values and longitudinal trajectories. Our main finding was that, although baseline PNI was not independently associated with 28-day mortality after adjustment for illness severity, longitudinal PNI during the first week of ICU stay remained significantly associated with mortality risk. In the joint model, each 10-point increase in current PNI was associated with a 16.9% lower hazard of 28-day mortality, independent of the covariates included in the model. Together, these findings suggest that serial assessment of PNI may provide additional prognostic information beyond a single admission measurement. The limited independent prognostic value of baseline PNI may reflect the fact that admission values in sepsis are strongly influenced by acute physiological disturbances [18]. The capillary leak that characterizes early sepsis, together with aggressive early fluid resuscitation, may lead to rapid hemodilution and albumin extravasation [19, 20]. Consequently, a low admission PNI may reflect not only the patient’s underlying condition but also the immediate impact of the acute inflammatory response and early therapeutic interventions. In this context, the prognostic information carried by PNI may be more apparent in longitudinal analyses, which better capture the patient’s subsequent clinical course. Our primary joint model used a current-value parameterization and indicated that the instantaneous PNI level was significantly associated with mortality risk. However, patients with different baseline values and opposite temporal trajectories could theoretically present similar current PNI levels. To address this, we fitted an alternative parameterization that separated baseline PNI from subsequent change over time (ΔPNI). The finding that ΔPNI was independently associated with survival indicates that the prognostic relevance of dynamic PNI is not solely attributable to its absolute level at any given moment. Instead, longitudinal change from baseline may also contribute meaningful prognostic information. These findings further support the clinical relevance of serial PNI assessment during ICU stay. We hypothesize that the prognostic relevance of dynamic PNI may relate to its ability to reflect two broad pathophysiological domains: vascular integrity and immune status [21, 22]. Although serum albumin has traditionally been regarded as a nutritional marker, its kinetics in acute illness are also influenced by endothelial barrier function [23]. In non-survivors, it is biologically plausible that persistent inflammation, endothelial glycocalyx degradation, and capillary leakage contribute to a flat or declining albumin trajectory [24, 25]. Conversely, improving albumin levels may accompany partial resolution of systemic inflammation and restoration of vascular homeostasis [26]. At the same time, the lymphocyte component may provide complementary information regarding the host immune state [27], including sepsis-associated immunosuppression related to lymphocyte apoptosis [28]. It is therefore plausible that declining PNI trajectories partly reflect persistent lymphocyte depletion and ongoing physiological stress. Taken together, these considerations suggest that PNI may serve as a composite clinical indicator integrating albumin- and lymphocyte-related information. However, our study did not include direct measurements of endothelial injury (e.g., Syndecan-1) or functional immune markers (e.g., mHLA-DR expression). Accordingly, these mechanistic interpretations should be regarded as hypothesis-generating rather than directly demonstrated by our data. At present, dynamic PNI is better interpreted as an integrated clinical surrogate of evolving physiological status that warrants further mechanistic investigation. These findings support the potential value of moving from static risk assessment toward dynamic monitoring in ICU patients [29–31]. In our analyses, dynamic PNI remained associated with mortality after adjustment for established severity indices, including APACHE II and SOFA. While SOFA remains a core tool for evaluating sepsis-related organ dysfunction [32], it primarily reflects the extent of organ failure rather than longitudinal changes in albumin- and lymphocyte-related physiology. In this context, serial PNI assessment may provide complementary prognostic information not fully captured by conventional severity scores alone [33]–[35]. From a clinical perspective, these results support further evaluation of serial PNI assessment as a pragmatic monitoring tool during ICU stay. A persistently low or declining PNI trajectory may help identify patients with an unfavorable clinical course and may justify closer reassessment of treatment response. However, the specific clinical actions triggered by such a pattern, and whether PNI-guided management can improve outcomes, require prospective validation before routine implementation can be recommended [36, 37]. A key methodological strength of this study is the use of joint modeling. Unlike conventional time-dependent Cox models alone, joint modeling explicitly links the longitudinal biomarker process with the survival process and helps account for informative dropout due to death or discharge. This framework may provide a more appropriate estimate of biomarker trajectories in critically ill populations with substantial early mortality. However, several limitations should be acknowledged. First, as a single-center retrospective study, our findings may have been influenced by institution-specific practices, such as fluid resuscitation protocols and nutritional support strategies, which may limit generalizability. Therefore, multicenter external validation across different geographical regions and critical care settings, including medical and surgical ICUs, is needed. Second, because inclusion required paired baseline measurements of serum albumin and total lymphocyte count within 24 h of ICU admission, selection bias cannot be completely excluded. Although included and excluded patients were broadly comparable with respect to the observed baseline variables, residual bias related to unmeasured factors or missing data mechanisms may still have influenced the results. Third, although we adjusted for major confounders, detailed fluid resuscitation volumes were not fully controlled, which may have affected early albumin kinetics and thus PNI values. Fourth, our study did not include direct biomarkers of endothelial injury or immune dysfunction. Therefore, the biological interpretation of dynamic PNI should be considered hypothesis-generating rather than mechanistically proven. Finally, the observational design precludes causal inference. It remains uncertain whether a declining PNI trajectory is merely a marker of disease progression or a modifiable determinant of poor outcomes. Future prospective studies should evaluate whether interventions guided by PNI trajectories can improve clinical outcomes. Conclusion In conclusion, longitudinal assessment of the Prognostic Nutritional Index (PNI) may provide additional prognostic information beyond baseline measurements in patients with sepsis. While admission values are influenced by acute illness severity and early resuscitation, serial changes in PNI may better reflect the evolving clinical course. In particular, a persistently low or declining PNI trajectory may help identify patients with a less favorable clinical course. However, the mechanistic basis and clinical utility of serial PNI assessment require further prospective validation. Supplementary Information Supplementary Material 1 Supplementary Material 1