Comparison of SOFA score alone versus SOFA score with interleukin-6 for outcome prediction in ICU patients with sepsis: a prospective observational study

Introduction

Sepsis remains a life-threatening condition characterized by organ dysfunction resulting from a dysregulated host response to infection, with persistently high mortality rates despite advances in critical care (1). Early identification of patients at highest risk of poor outcomes is crucial, as timely intervention has been shown to improve clinical outcomes (2).

The Sequential Organ Failure Assessment (SOFA) score, endorsed by the Sepsis-3 definition, provides an objective measure of organ dysfunction across six systems (3). Although originally designed to describe the degree of organ failure rather than predict outcomes, SOFA has been widely adopted for prognostication in critically ill patients (4). However, SOFA has inherent limitations: it relies solely on physiological parameters obtained at specific timepoints and does not account for the underlying inflammatory burden, immune dysregulation, or patient-specific factors such as age and comorbidities that influence outcomes (5).

A meta-analysis by Fleischmann-Struzek et al. reported an estimated mortality rate of 26.7% among septic patients, along with a 46% increase in incidence after 2008, emphasizing the growing clinical impact of this condition (6).

The heterogeneity and complexity of sepsis pose significant challenges in identifying high-risk patients who may benefit from closer monitoring and aggressive therapy (7). Prognosis and clinical results are influenced by several host- and pathogen-related factors, such as age, comorbidities, source of disease, and individual immune responses, particularly reflected through circulating biomarkers such as cytokines and interleukins (8). The accurate prediction of outcomes is therefore essential for guiding therapeutic decisions and resource allocation and for informing patients and their families regarding prognosis. Furthermore, assessment of illness severity in critically ill patients serves as an important indicator of ICU performance and quality of care (9).

The SOFA score was originally developed following a consensus meeting organized by the European Society of Intensive Care Medicine in Paris in 1994, with the aim of quantitatively and objectively describing the degree of organ dysfunction or failure over time in critically ill patients. Although it was not initially designed to predict mortality, its utility in assessing acute morbidity and identifying critical illness led to its adoption as a sequential assessment tool (10). However, commonly used predictive scoring systems like sequential organ failure assessment (SOFA), acute physiology and chronic health evaluation (APACHE), simplified acute physiology score (SAPS), and mortality probability mode (MPM) primarily rely on physiological parameters obtained on Day 0 of admission to the ICU and do not adequately account for factors such as age, comorbidities, or evolving disease processes that may influence outcomes beyond this period (11).

To strengthen the prognostic accuracy of SOFA, various studies have explored its integration with biomarkers that reflect the underlying inflammatory and immune responses in sepsis. Among these, interleukin-6 (IL-6) has emerged as a promising candidate. Critically, Shimazui et al. demonstrated that a patient’s IL-6 level rises earlier than procalcitonin (PCT) and C-reactive protein (CRP) following a septic insult and peaks sooner than the SOFA score, suggesting its potential as an early prognostic indicator (12).

Interleukin-6 is a pro-inflammatory cytokine synthesized by T-lymphocytes, monocytes, fibroblasts, and endothelial cells and plays a critical role in the acute-phase inflammatory response in sepsis (13). IL-6 is released as a result of tissue injury or inflammatory stimuli and exerts both local and systemic effects, with serum concentrations rising following surgery, trauma, and critical illness. The integration of IL-6 with SOFA may therefore offer distinct advantages: IL-6 provides an early, sensitive measure of the host inflammatory response, while SOFA captures the consequent organ dysfunction over time. Their combination could potentially enhance prognostic accuracy by addressing the temporal disconnect between the initial inflammatory cascade and the subsequent development of organ failure (14).

Despite increasing evidence supporting the individual prognostic significance of SOFA and IL-6, there remains a paucity of studies evaluating the diagnostic precision of their combined use for mortality risk assessment in septic patients. This hypothesis-generating study was therefore undertaken with a primary objective of evaluating the prognostic accuracy of the SOFA score and IL-6 individually and in combination for outcome assessment in patients with sepsis. The secondary objectives were to evaluate the temporal trends of SOFA scores and IL-6 levels among survivors and non-survivors during the first 72 h of ICU admission, determine the optimal cut-off values for SOFA and IL-6 at different timepoints for mortality prediction, and assess the sensitivity, specificity, and predictive values of the combined score in comparison with the SOFA score alone.

Materials and methods

The study was conducted after obtaining approval from the Institutional Ethics Committee (F.1/IEC/MAMC/(82/10/2020/No.21). The study was conducted prospectively over a period from May 2021 to August 2022. Written informed consent was obtained from the patients (when possible) or their relatives. Adult patients with underlying sepsis between the age group of 18–50 years with an anticipated length of ICU stay >72 h were included in the study. Patients with acute coronary syndrome, terminal cancers, acute burn injuries, on mechanical ventilation after elective surgeries, coronavirus disease 2019 (COVID-19), and those who were transferred from other hospitals’ ICUs were excluded from the study.

The 2016 Surviving Sepsis guidelines (Sepsis-3) were followed for the diagnosis of sepsis in hospitalized individuals. The patients’ SOFA score and interleukin-6 level were obtained from the ICU scoring data sheet at the time of ICU admission (0 h) and were subsequently recorded at 24-h intervals for a total duration of 72 h. The pre-sedation Glasgow Coma Scale (GCS) score was used to assess the neurological status of the patients under sedation. All the patients in the ICU were subsequently followed up for 15 days. For patients who were discharged before 15 days, telephonic communication about their welfare was maintained for 15 days. The 15-day endpoint was chosen as it represents sepsis-attributable mortality more directly than 30-day outcomes, minimizes loss to follow-up, and aligns with the period during which SOFA demonstrates optimal discriminative ability. The predicted outcome of survivor or non-survivor based on the scores at 0, 24, 48, and 72 h (Days 0, 1, 2, and 3, respectively) was determined and correlated with the actual outcome on Day 15 (Figure 1).

Assessment of components of the SOFA score

Each component of the SOFA score was systematically evaluated. Neurological function was assessed using the GCS, with scores recorded after discontinuing sedative infusions in mechanically ventilated patients to ensure accuracy. The respiratory component was determined by the lowest arterial oxygen partial pressure/fraction of inspired oxygen (paO₂/FiO₂) ratio from arterial blood gas analyses within each 24-h period. Cardiovascular scoring was centered on the highest level of vasopressor or inotropic support required during the assessment window. Renal function was evaluated using serum creatinine and 24-h urine output; patients receiving renal replacement therapy were automatically assigned a score of 4. The coagulation sub-score was calculated from the lowest platelet count recorded in the preceding 24 h.

Measurement of the SOFA score and interleukin 6 level

The SOFA score for each patient was calculated using the scoring system proposed by Yang et al. (15). We estimated the baseline SOFA score upon admission (0 h) and at every 24 h subsequently till 72 h (Days 0, 1, 2, and 3, respectively). Similarly, the serum IL-6 values were obtained at the given timepoints, as in the study conducted by Bhandari et al. (16) (Table 1). Although Bhandari et al. studied patients with COVID-19, the IL-6 thresholds are conservative when applied to sepsis, as IL-6 levels in septic shock are significantly higher than in COVID-19 (17). This approach therefore provides a stringent cut-off that enhances specificity without compromising sensitivity.

Table 1. Scoring system for interleukin-6 level.

Scoring system adapted from Bhandari et al. (16).

A categorical approach was preferred over raw IL-6 values to create a clinically simple combined score that can be calculated at the bedside without statistical software. Categorization into low, intermediate, and high thresholds reduces the influence of extreme outliers while preserving prognostic information.

To equalize the contribution of both parameters, a combined score was generated by modifying the SOFA score and adding the IL-6 score, following the method described by Yang et al. (15). A multiplication factor of 0.1 was applied to scale the SOFA score (range 0–24) to a range comparable to the IL-6 score (0–2). This prevents the SOFA score from dominating the combined score solely due to its larger numerical range and is a standard scaling approach when combining variables measured on different scales.

Modified SOFA score = SOFA score×0.1 (minimum score of 0, maximum score of 2.4)

Combined score = Modified SOFA + IL-6 score (minimum score of 0, maximum score of 4.4)

A systematic review and meta-analysis by de Grooth et al. (11), which pooled data from 43 randomized controlled trials involving 13,784 critically ill patients, demonstrated that baseline SOFA scores consistently predict mortality, with non-survivors having mean SOFA scores 3–4 points higher than survivors. Based on an expected sensitivity of 85% for SOFA in predicting mortality, the sample size was calculated at a 95% confidence level (Z_{₁₋*α*/₂} = 1.96). Considering the exploratory nature of the study and feasibility constraints in a single-center ICU setting, a margin of error of 15% (d = 0.15) was used, yielding a minimum sample size of 26 patients. To improve reliability and account for possible exclusions or incomplete follow-up, 40 participants were ultimately enrolled.

Statistical analysis was performed using SPSS for Windows, version 25.0 (SPSS Inc., Chicago, IL, USA) and R version 4.2 (R Foundation for Statistical Computing, Vienna, Austria). Continuous variables are expressed as mean ± standard deviation or median with interquartile range, depending on data distribution. Categorical variables, including gender, diagnosis, and mortality, are presented as frequencies and percentages and were compared using the chi-square test or Fisher's exact test, as appropriate.

Patients were categorized into groups, survivors and non-survivors, based on their outcome at the end of the study period. Demographic and clinical variables, including SOFA scores and IL-6 levels at different timepoints, were compared between the two groups. For ordinal data (SOFA score and combined score), group comparisons were performed using the Mann–Whitney U test. For continuous variables (IL-6 level), Student's t-test was applied for normally distributed data, while the Mann–Whitney U test was used for non-normally distributed data.

Receiver operating characteristic (ROC) curve analysis was undertaken to determine the ability of the SOFA score and the combined SOFA–IL-6 score to predict mortality. The area under the curve (AUC), sensitivity, and specificity were calculated to assess diagnostic accuracy. A comparison of AUCs between SOFA alone and the combined score was performed using bootstrap confidence intervals. Statistical significance was defined as a p-value less than 0.05 for all analyses.

Internal validation using bootstrap resampling was performed to assess potential model overfitting, as recommended for prediction models developed in small-to-moderate sample sizes. A total of 1,000 bootstrap samples were generated with replacement from the original dataset (n = 40). For each bootstrap sample, the combined score model was refit and the AUC recalculated. The optimism was calculated as the difference between the apparent AUC and the mean bootstrap AUC, and the optimism-corrected AUC was derived by subtracting the optimism from the apparent AUC. This approach provides a more reliable estimate of model performance in future patients. Bootstrap analyses were conducted using the pROC and boot packages in R.

Results

A total of 40 critically ill patients were enrolled, all of whom completed the study. At the end of the 15-day follow-up period, 24 patients (60%) survived, while 16 patients (40%) died.

Among survivors, the mean age was 35.62 ± 11.16 years (range: 18–50 years), whereas in the non-survivor group, the mean age was 45.75 ± 4.67 years (range: 35–50 years). In both groups, the gender distribution was skewed toward males, accounting for 54.1% of survivors and 81.25% of non-survivors. Medical comorbidities were present in 16 patients (40%), with diabetes mellitus observed in 6 patients (15%), cardiac diseases (including hypertension) and kidney disease in 9 patients (22.50%) each, and neurological diseases in 7 patients (17.5%). All patients had respiratory involvement (some patients had more than one comorbidity).

Both medical and surgical patients with underlying sepsis requiring ICU care were included. Among the survivors, medical patients constituted 29.2%, while surgical patients accounted for 70.8%. In contrast, among the non-survivors, medical conditions were present in 56.2% and surgical conditions in 43.8%. Among the surgical cohort, the largest subgroup consisted of nine patients (22.5%) who underwent emergency exploratory laparotomy, followed by six patients (15%) with complicated gallstone disease and two patients (5%) with postoperative cesarean section complications. Additionally, three patients (7.5%) with a history of trauma required ICU admission. Infection source and microbiological data were not collected as part of this study protocol, as the primary focus was on physiological and biomarker parameters.

The mean duration of ICU stay was significantly longer among the survivors (13.29 ± 7.68 days) compared with the non-survivors (7.38 ± 2.92 days), with this difference reaching statistical significance (p = 0.006).

SOFA score comparison

The median SOFA score at ICU admission (Day 0) was higher among the non-survivors [13 (IQR 10–15)] compared with the survivors [10 (IQR 8–12)]. Over time, a decreasing trend in SOFA scores was observed among the survivors, whereas an increasing trend was noted among the non-survivors (Table 2). The difference in SOFA scores between the two groups was statistically significant at all timepoints (p < 0.05).

Table 2. Comparison of SOFA score between the survivors and non-survivors.

Data are presented as median (interquartile range). *Group comparisons performed using the Mann–Whitney U test (non-parametric) due to the ordinal nature of the SOFA scores.

IL-6 level comparison

Within the non-survivor group, no consistent temporal trend in IL-6 levels was observed, and the differences across timepoints were not statistically significant. However, mean IL-6 levels were significantly higher in non-survivors compared with survivors at all timepoints (p < 0.05; Table 3).

Table 3. Comparison of IL-6 level (pg/mL) between the survivors and non-survivors.

IL-6, interleukin-6; SD, standard deviation. Data are presented as mean ± SD. Comparison between survivors and non-survivors was performed using the independent Student's t-test. Bold p-values indicate statistical significance (p < 0.05).

Combined SOFA–IL-6 score comparison

The combined score, calculated as (SOFA × 0.1) + IL-6 score (0–2), was significantly higher in the non-survivors compared to the survivors at all timepoints (p < 0.001). The non-survivors showed a progressive increase in combined scores over time, while the survivors demonstrated a declining trend (Table 4).

Table 4. Comparison of the combined SOFA–IL-6 score between the survivors and non-survivors.

Mann–Whitney U test.

ROC analysis of SOFA and IL-6

The discriminative ability of the SOFA score and IL-6 level for predicting mortality was assessed using ROC curve analysis (Figure 2). The AUC for SOFA increased progressively over time, with Day 3 demonstrating the best apparent performance, i.e., an apparent AUC of 0.986 (95% CI: 0.958–0.999), sensitivity of 81.2%, and specificity of 100%. The AUC for SOFA at admission (Day 0) was 0.758 (95% CI: 0.598–0.918), indicating fair discriminative ability. On Day 1, the AUC was 0.755 (95% CI: 0.605–0.905), and on Day 2, it was 0.861 (95% CI: 0.747–0.974) (Table 5).

Table 5. Bootstrap-corrected AUCs for SOFA alone and the combined score.

Bootstrap validation performed with 1,000 resamples. Optimism = mean bootstrap AUC−apparent AUC. Corrected AUC = apparent AUC−optimism. 95% CI represents the bootstrap percentile confidence interval for the corrected AUC.

Baseline IL-6 levels demonstrated fair discriminative ability with an AUC of 0.758 (95% CI = 0.611–0.904). On Day 1, IL-6 showed similar performance with an AUC of 0.786 (95% CI = 0.645–0.928). Improved accuracy was observed on Day 2, with an AUC of 0.822 (95% CI = 0.680–0.963). The highest discriminative performance was noted on Day 3, with an apparent AUC of 0.993 (95% CI = 0.977–0.999), sensitivity of 100%, and specificity of 87.5% (Figure 2).

However, bootstrap internal validation with 1,000 resamples revealed modest overfitting in both models. Optimism values ranged from 0.044 to 0.082 across models and timepoints. The optimism-corrected AUC for SOFA alone on Day 3 was 0.908 (95% CI: 0.826–0.990). The optimism-corrected AUC for the combined score on Day 3 was 0.908 (95% CI: 0.824–0.992), indicating that both models perform equivalently after correction for overfitting. On Days 0–2, the combined score demonstrated marginally higher corrected AUCs compared to SOFA alone, as follows: Day 0: 0.734 vs. 0.710, Day 1: 0.751 vs. 0.703, and Day 2: 0.830 vs. 0.817, representing improvements of +0.024 to +0.048 (Table 5).

Prognostic performance of SOFA alone vs. the combined SOFA–IL-6 score

The primary objective of this study was to evaluate whether combining IL-6 with SOFA improves prognostic accuracy. A combined score was therefore created by scaling the SOFA score (multiplied by 0.1 to achieve a range of 0–2.4, comparable to the IL-6 score of 0–2) and adding the IL-6 score derived from categorized IL-6 levels (0 for <7 pg/mL, 1 for 7–70 pg/mL, and 2 for >70 pg/mL) (Table 1). The combined score was calculated as: Combined Score = (SOFA × 0.1) + IL-6 score.

After bootstrap internal validation, the combined score demonstrated corrected AUCs of 0.734 (Day 0), 0.751 (Day 1), 0.830 (Day 2), and 0.908 (Day 3). While the apparent AUCs suggested the superiority of the combined score, the corrected estimates revealed that on Day 3, both SOFA alone and the combined score performed equivalently (corrected AUC of 0.908 for both). On Days 0–2, the combined score showed modest improvements over SOFA alone, with corrected AUC differences of +0.024 to +0.048 (Table 5, Figure 3).

Optimal cut-off values for classifying patients as high-risk were determined from ROC analysis by selecting the points that maximized both sensitivity and specificity. For clinical applicability, SOFA cut-offs were rounded to the nearest whole number. Patients were classified as positive if they met both criteria as follows: SOFA ≥ 11 and IL-6 ≥ 85.5 pg/mL on Day 0; SOFA ≥ 11 and IL-6 ≥ 70.5 pg/mL on Day 1; SOFA ≥ 10 and IL-6 ≥ 72.0 pg/mL on Day 2; and SOFA ≥ 10 and IL-6 ≥ 81.5 pg/mL on Day 3. All the remaining patients were classified as negative.

Predictive values of SOFA alone and the combined SOFA–IL-6 score

To further evaluate the clinical utility of both scoring systems, 2 × 2 contingency tables were constructed using the optimal cut-off values derived from ROC analysis for Day 3 (the timepoint with the highest discriminative ability). The positive predictive value (PPV), negative predictive value (NPV), and diagnostic accuracy were calculated for both SOFA alone and the combined score (Table 6). It should be noted that these predictive values are based on apparent performance and are likely optimistic. Bootstrap-corrected estimates are reported in Table 5.

Table 6. Comparison of the predictive values of SOFA alone vs. the combined SOFA–IL-6 score on Day 3 (apparent performance).

Combined score cut-off derived from ROC analysis (SOFA × 0.1 + IL-6 score) with optimal Youden index.

Discussion

The prognostic significance of clinical scoring systems and biochemical markers in sepsis has been extensively investigated. While various studies have focused on assessing disease severity using established scoring systems, relatively few have explored their capacity to predict outcomes, particularly short-term mortality. The current study was designed to assess the prognostic performance of the SOFA score when integrated with IL-6 levels for predicting 15-day mortality outcomes in patients with clinically suspected sepsis.

In the current study, SOFA alone demonstrated good discriminative ability, with an apparent AUC of 0.986 and an optimism-corrected AUC of 0.908 on Day 3. These findings are broadly consistent with previously reported AUC values ranging from 0.722 to 0.879 for SOFA-based mortality prediction in critically ill patients (18–25). Notably, the majority of these studies did not evaluate the temporal trend of SOFA scores. In contrast, our longitudinal assessment demonstrated that a decreasing SOFA score over time was associated with reduced mortality, highlighting the importance of dynamic evaluation rather than isolated measurements.

The moderate prognostic accuracy of SOFA observed in our study may be attributable to several factors. Mortality in septic ICU patients is influenced not only by the extent of organ dysfunction but also by patient-related variables such as age and comorbidities, as well as infection-related factors including pathogen virulence, site of infection, and microbial profile. These parameters are not incorporated into the SOFA score, which may partially explain its limited discriminatory power for mortality prediction. Supporting this observation, Lee et al. evaluated a complemented SOFA score by incorporating age and comorbidity indices and demonstrated significantly improved prognostic accuracy compared to SOFA alone (AUC 0.812 vs. 0.743, p < 0.001) (26).

There were significant differences in baseline age and gender between the survivors and non-survivors. The non-survivors were older (45.75 vs. 35.62 years) and had a higher proportion of males (81.25% vs. 54.1%). Older age and male gender are known risk factors for poorer outcomes in sepsis. These imbalances may have influenced the prognostic performance of both SOFA and IL-6. Future larger studies should adjust for age and gender in multivariable models to assess the independent prognostic value of these scores.

Biomarkers such as IL-6, PCT, and CRP, when included with clinical scoring systems, may enhance prognostic precision by reflecting the underlying inflammatory burden, immune dysregulation, endothelial injury, and cellular dysfunction—pathophysiological processes not fully captured by organ-based scores alone (10, 13). Song et al. emphasized the additive value of biomarkers in complementing SOFA-based risk stratification in critically ill patients (13).

In the current study, IL-6 independently demonstrated moderate prognostic accuracy at ICU admission, with an apparent AUROC of 0.758 at a cut-off value ≥85.5, yielding a sensitivity of 81.2% and specificity of 62.5%. This performance was comparable to, and slightly greater than, that reported by Jiang et al. (AUROC of 0.712) (22). Importantly, the prognostic precision of IL-6 improved over subsequent days. The apparent AUROC values for Days 1, 2, and 3 were 0.786 (cut-off ≥70.5; sensitivity of 93.8%, specificity of 62.5%), 0.822 (cut-off ≥72.0; sensitivity of 81.2%, specificity of 83.3%), and 0.993 (cut-off ≥81.5; sensitivity of 93.8%, specificity of 100%), respectively, demonstrating a progressive increase in discriminative power.

These findings are consistent with Shimazui et al., who demonstrated that IL-6 more accurately reflected organ dysfunction severity compared to PCT and CRP, with significant differences across SOFA score categories (p < 0.0001) (12). They also observed that IL-6 peaked earlier than both other biomarkers and the SOFA score itself, supporting its role as an early indicator of disease severity, a finding that aligns with the progressive improvement in IL-6's prognostic performance observed in our study.

Early identification of high-risk septic patients is critical, as timely prognostication may facilitate earlier escalation of care and optimization of therapeutic interventions. The principal finding of this study is that combining IL-6 with SOFA demonstrated modest improvements in prognostic accuracy on Days 0–2 compared to SOFA alone, with corrected AUC differences of +0.024 to +0.048. However, after bootstrap internal validation to correct for overfitting, the combined score and SOFA alone performed equivalently on Day 3 (corrected AUC of 0.908 for both).

The optimism values (0.048–0.082) confirm that overfitting, while present, was modest. The reduction from the apparent Day 3 AUC of 0.990 to the corrected AUC of 0.908 highlights the importance of internal validation in small-sample prediction studies. These corrected estimates provide a more realistic expectation of model performance in future patients.

While the apparent (uncorrected) performance suggested that the combined score eliminated false negatives entirely, this finding was likely inflated due to overfitting. After correction, both models demonstrated equivalent performance on Day 3. The modest improvements observed on Days 0–2 suggest potential for earlier risk stratification, though this requires validation.

The ability to identify all patients at risk of death by Day 3 has important clinical applications. First, it allows for timely re-evaluation of treatment strategies in patients who are not improving as expected. Second, it facilitates informed discussions with families regarding prognosis. Third, it may guide resource allocation, including consideration of advanced therapies or transfer to specialized centers for high-risk patients identified by the combined score. The 100% NPV of the combined SOFA–IL-6 score on Day 3 (Table 6) is particularly valuable, as it reliably identifies survivors and may prevent unnecessary interventions in this group. The practical implementation of this combined score is feasible in most ICU settings. IL-6 assays are increasingly available, and the calculation only requires simple arithmetic.

Limitations

The present study also had certain limitations. It was conducted at a single center with a modest sample size, which limits the precision of our estimates and necessitated internal validation. A multicenter design was not feasible during the study period (May 2021–August 2022) due to COVID-19-related restrictions on ICU research. This hypothesis-generating study was therefore designed to inform future larger-scale validation studies. While bootstrap resampling was performed to correct for overfitting, external validation in an independent cohort remains essential before any clinical implementation. The corrected AUC of 0.908 on Day 3, while excellent, is lower than the apparent AUC of 0.990, and further shrinkage may occur in external validation. Our inclusion of only patients aged 18–50 years substantially limits generalizability to older patients with sepsis. This age restriction was chosen to minimize confounding from age-related comorbidities in this hypothesis-generating study. A subgroup analysis comparing medical and surgical patients was not feasible due to the small sample size. The study cohort was heterogeneous, and only short-term (15-day) mortality was evaluated. The small sample size also precluded more complex multivariable models. Additionally, although several emerging biomarkers have been investigated for prognostication in sepsis, only IL-6 was evaluated in combination with SOFA in this study. Future large-scale, prospective, multicenter studies are warranted to validate these findings and to explore whether alternative or additional biomarker combinations may further enhance prognostic accuracy in sepsis.

Conclusion

The combined SOFA–IL-6 score showed moderate discriminative ability for 15-day mortality that was comparable to SOFA alone by the third day of ICU admission after internal validation, with only modest improvements during the earlier period following admission. These findings are exploratory and do not support the superiority of the combined score. External validation in larger independent cohorts is required before clinical application.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Marco Fiore, Università degli Studi della Campania ‘Luigi Vanvitelli', Italy

Reviewed by: Navid Shafigh, Shahid Beheshti University, Iran

Resiana Karnina, University of Indonesia, Indonesia