Early Prediction of In-Hospital Mortality in Patients with Acute Infections: Development of the Acute Severity Infection Score (ASIs)

Abstract

Introduction

Early prognostic stratification in patients hospitalized for acute infections is a major clinical challenge. Existing tools, such as the Sequential Organ Failure Assessment (SOFA) score and Charlson Comorbidity Index (CCI), were not specifically developed for this purpose.

Objectives

We aimed to design a novel multidimensional prognostic score, the Acute Severity Infection score (ASIs), to predict in-hospital mortality using routinely available clinical data.

Methods

This retrospective cohort study included 149 adults admitted with acute infections to an internal medicine unit between January 2023 and December 2024. In-hospital all-cause mortality was the primary outcome. Demographic, clinical and laboratory variables obtained within 12 h of admission were analyzed. Variables significantly associated with mortality in both univariate and multivariate regression were incorporated into the ASIs, which ranges from 0 to 7 points. Its performance was compared to SOFA and CCI using ROC curve and Cox regression models.

Results

In-hospital mortality occurred in 25.5% of patients. Five variables were independently associated with mortality: lactate ≥ 2.2 mmol/l, frailty composite (confined to bed status, long-term oxygen therapy or advanced malignancy), hemodynamic instability or need for non-invasive ventilation, age ≥ 79.5 years and symptom onset ≥ 3.5 days before admission. ASIs showed the highest discriminative ability (AUC = 0.883) compared to SOFA (AUC = 0.612) and CCI (AUC = 0.742). In multivariate models including all three scores, only ASIs retained independent prognostic significance.

Conclusions

The ASIs is a simple tool for early prognostic stratification of patients hospitalized with acute infections. It outperforms existing scores and may enhance clinical decision-making in real-world medical settings.

Key Summary Points

Early prognostic stratification in patients hospitalized for acute infections is a major clinical challenge. Identifying patients at high risk of mortality has high clinical relevance and may be crucial in guiding therapeutic management.

The study aimed to design a novel multidimensional prognostic score (Acute Severity Infection score—ASIs) to predict in-hospital mortality using routinely available clinical data.

ASIs proved to be a simple yet powerful tool for early prognostic stratification of patients hospitalized with acute infections.

It outperformed Sequential Organ Failure Assessment (SOFA) and Charlson Comorbidity Index (CCI) in predicting in-hospital mortality (AUC = 0.88; standardized HR = 3.4).

Introduction

Acute infections remain one of the leading causes of hospitalization and, when complicated by sepsis and septic shock, are associated with high morbidity and mortality rates. The early identification of patients at higher risk of mortality is therefore crucial to optimizing diagnostic and therapeutic pathways and improving clinical outcomes [1]. However, identifying high-risk patients remains challenging, requiring a comprehensive approach that integrates medical history, physical examination and laboratory findings while also considering the underlying pathophysiologic mechanisms [2]. Although clinical judgment is implicitly used in the management of critically ill patients, no systematic tool is currently available to objectify and standardize its application [3].

The current literature highlights the prognostic role of serum lactate in sepsis, not only as a marker of tissue hypoperfusion but also as an indicator of cellular metabolic stress [4]. Similarly, C-reactive protein (CRP) and procalcitonin (PCT), despite only moderate diagnostic accuracy, are well-established tools for infection assessment and treatment monitoring [5]. However, an approach based on single biomarkers is limited, failing to account for the complexity of septic disease and the role of patient-specific baseline characteristics [6].

Therefore, integrating these variables into a single multidimensional tool could provide a more accurate assessment of in-hospital mortality risk [7]. Currently, the primary prognostic tool used for risk stratification in patients with sepsis is the Sequential Organ Failure Assessment (SOFA) score, while the most relevant score for assessing patient comorbidities and predicting long-term mortality is the Charlson Comorbidity Index (CCI). The SOFA score is the international reference for assessing organ dysfunction in critically ill patients. However, it does not incorporate key prognostic factors, such as inflammatory/metabolic status and pre-existing comorbidities, which may significantly impact patient outcomes [8, 9]. Moreover, since it was originally validated in an intensive care unit (ICU) setting [10], its applicability in general medical wards remains limited. Conversely, the CCI is widely employed to quantify the comorbidity burden but was not specifically designed for acute infections and does not account for dynamic parameters of clinical progression [11]. In this study, the CCI was used as a complementary comparator to represent the dimension of chronic comorbidity burden alongside acute illness severity (SOFA score).

In this context, the present study aims at developing a multidimensional prognostic score based on three main domains: patient comorbidities and vulnerability, disease severity objective clinical signs and inflammatory-metabolic biomarkers. The first domain considers chronic conditions with a significant prognostic impact, such as cardiovascular disease, advanced liver disease, chronic respiratory failure, advanced-stage malignancy and end-stage renal disease, as well as frailty conditions such as confined to bed status [12].

The second domain focuses on the need for organ support, such as non-invasive mechanical ventilation (NIV) or vasopressor therapy, as a critical indicator of disease severity. In addition to these interventions, it includes easily identifiable clinical signs, such as peripheral edema and ascites, which may reflect fluid distribution abnormalities and influence outcomes in patients with sepsis [13].

Finally, the third domain includes inflammatory and metabolic stress biomarkers, such as CRP, PCT and serum lactate, which are widely used in clinical practice for risk stratification and therapeutic monitoring [14].

The ultimate goal is to propose a novel, easy-to-use and cost-effective score capable of improving the prognostic assessment of patients with acute infections by more accurately predicting in-hospital mortality. This tool is specifically designed for use in medical wards, providing clinicians with a practical and reliable instrument to support personalized and timely patient management.

Methods

Study Design and Population

This was a retrospective cohort study conducted at the Internal Medicine Unit of the "SS Annunziata" Hospital in Taranto, Italy. This Unit includes a dedicated high-dependency area equipped with continuous vital sign monitoring and NIV and the capacity to initiate vasopressor therapy when required. Patients are managed in this setting under the supervision of internal medicine specialists, with additional support from intensivists when indicated. Those requiring invasive mechanical ventilation or advanced hemodynamic support are transferred to the ICU.

We included adult patients (≥ 18 years) admitted between January 2023 and December 2024 whose discharge diagnosis confirmed an acute infectious disease (AID) as the principal reason for hospitalization (e.g., pneumonia, cholecystitis, sepsis, septic shock, colitis, cholangitis, urinary tract infection, pyelonephritis, diverticulitis or respiratory tract infection).

To ensure diagnostic consistency, only cases in which the infectious disease diagnosis was supported by clinical, laboratory, microbiologic and/or imaging findings, according to current international guidelines, were included.

Patients were excluded if they were pregnant or breastfeeding, they had incomplete medical documentation, the available data referred to clinical or laboratory parameters obtained > 12 h after admission in our unit or they had received antibiotic therapy or fluid resuscitation (> 1000 ml crystalloids) in the emergency department (ED) before admission. In this regard, it should be highlighted that in our hospital, the ED typically performs rapid initial assessment and triage of patients with suspected infections but does not routinely administer intravenous antibiotics or large-volume fluid resuscitation before admission to our Internal Medicine Unit. The administration of early treatment in the ED occurs only in few cases because of institutional protocols that prioritize rapid transfer to our Unit for initiation of therapy and definitive management. Therefore, the exclusion of patients who received early antibiotic therapy or > 1000 ml fluid resuscitation in the ED aimed to ensure homogeneity of baseline measurements rather than to select for less severe cases.

Mortality was assessed as in-hospital all-cause mortality and further categorized into early (≤ 7 days from admission) and late (between 8 and 28 days). Patients were considered survivors if they remained alive for at least 28 days following hospital admission.

Data Collection

Clinical and laboratory data were retrospectively extracted from electronic medical records, referring to the first 12 h after admission to the Internal Medicine Unit.

The following variables were systematically collected:

• Demographic and clinical history: age, gender, time from symptom onset to hospital admission and chronic comorbidities (e.g., cardiopathy, COPD, chronic respiratory failure, stroke, liver disease, chronic kidney disease, dialysis, bed confinement, peripheral vascular disease, dementia, connective tissue disorders, peptic ulcer, diabetes mellitus, solid and hematologic malignancies, chronic infections). The CCI was calculated for each patient.

  Specific operational definitions were adopted for clinical conditions that were potentially heterogeneous or prone to interpretation across medical records:

  • Advanced malignancy was defined as metastatic solid tumors or hematologic cancers no longer amenable to curative treatment in line with definitions used in palliative oncology literature [15];
  • Bed confinement referred to patients with persistent immobility due to severe physical disability, advanced neurologic disease or functional dependence requiring full assistance for bed-to-chair transfers [16];
  • Cardiopathy included chronic heart failure (NYHA class II–IV), a history of myocardial infarction or structural cardiac diseases associated with hemodynamic relevance [17];
  • Chronic respiratory failure was defined by the prescription of long-term oxygen therapy (LTOT) at home for resting hypoxemia (PaO₂ ≤ 55–60 mmHg in ambient air), in accordance with clinical guidelines [18].

• Laboratory parameters: CRP, PCT, lactate, creatinine, platelet count, bilirubin, arterial blood gas values.

• Clinical parameters: blood pressure, Glasgow Coma Scale (GCS), need for vasopressor therapy, requirement for NIV and infection site. Hemodynamic instability was defined as a mean arterial pressure (MAP) ≤ 65 mmHg or the need for vasopressors agents to maintain adequate perfusion, in accordance with Sepsis-3 definitions [3]. The SOFA score was calculated for each patient.

• Physical examination findings: presence of peripheral edema, ascites.

• Microbiologic findings: identification of pathogenic organisms (e.g., blood culture).

Outcomes

The primary endpoint was to assess the prognostic impact of clinical and laboratory variables on in-hospital mortality to create a predictive prognostic score based on the combination of these factors.

The secondary endpoint was to compare the predictive performance of the newly developed score with established prognostic scores such as the SOFA score and the CCI.

Sample Size Calculation

The required sample size was calculated based on the primary endpoint, using assumptions derived from Cox proportional hazards regression.

Assuming an expected in-hospital mortality rate of 20% (based on preliminary data and literature), a hazard ratio of 1.5 for at least one candidate prognostic variable, a two-sided alpha level of 0.05 and 90% power, a total of 140 patients was estimated to be necessary to detect a statistically significant difference in the risk of in-hospital mortality between survivors and non-survivors in our population.

Statistical Analysis

Continuous variables were expressed as mean ± standard deviation (SD) or median ± interquartile range (IQR) as appropriate, based on distribution. Categorical variables were reported as absolute frequencies and percentages.

Between-group comparisons were performed using Student’s t-test or Mann-Whitney U test for continuous variables and χ² test or Fisher’s exact test for categorical variables.

Variables that showed a statistically significant difference between survivors and non-survivors were tested in a ROC analysis. Continuous variables with an area under the curve (AUC) > 60% and p < 0.05 were dichotomized using the Youden index. Variables that did not reach statistical significance due to limited sample size, but held clinical relevance and biologic plausibility, were combined into composite variables and reassessed. The resulting variables were independently evaluated in binary logistic regression models corrected for sex and age, using overall in-hospital mortality as the dependent variable. A multivariable model including all selected predictors was then constructed to evaluate their independent contribution, and multicollinearity was assessed through the variance inflation factor (VIF). Based on AUC, confidence intervals, p-values, odds ratios and β coefficients, weighted scores were assigned to each variable, leading to the development of the ASIs score.

Finally, the independent prognostic value of the ASIs was assessed in univariate and multivariate Cox regression models, including SOFA and CCI as covariates. Hazard ratios were also standardized, using z-score transformation, to allow direct comparison across different scoring systems.

All statistical analyses were conducted using SPSS software (version 27.0.1.0, IBM Corp., Armonk, NY, USA). A two-tailed p value < 0.05 was considered statistically significant.

Ethical Considerations

The study was conducted in accordance with the general ethical principles for medical research involving human subjects of the Declaration of Helsinki [19]. The study protocol was formally approved by the Clinical Investigation Ethics Committee of the University of Bari (ID: Studio ASIs 2025, protocol number 2150/CEL, version 1.01, 19 March 2025). As a retrospective study based on anonymized clinical data, formal informed consent was waived by the ethics committee.

Results

Population Characteristics

A total of 149 patients, whose baseline characteristics are summarized in Table 1, were included in this study. The mean age was 76.5 years, and 55.7% were male. The median time from symptom onset to hospital admission was 3 days.

Table 1.

Baseline characteristics of the study population and comparison between male and female patients

Variable	Cohort (n = 149)	Male (n = 83)	Female (n = 66)	p value
Age, years (mean ± SD)	76.5 ± 12.2	76.4 ± 10.9	76.6 ± 13.7	0.902
Days from symptom onset, median (IQR)	3 (1–5)	3 (1–5)	3 (1–4)	0.932
CCI (median, IQR)	6 (5–7)	6 (5–8)	6 (5–7)	0.327
SOFA (median, IQR)	3 (2–5)	3 (2–5)	3 (2–5)	0.253
Hemodynamic instability n (%)	10 (6.7)	3 (3.6)	7 (10.6)	0.109
CRP, mg/l (median, IQR)	117.8 (51.1–199.8)	117.8 (57–200.6)	115.0 (50.8–202)	0.888
PCT. ng/ml (median, IQR)	1.67 (0.2–11.8)	1.2 (0.2–11.5)	1.9 (0.2–16.6)	0.438
Lactate, mmol/l (median, IQR)	1.8 (1.2–2.6)	1.8 (1.2–2.6)	1.8 (1.1–3.0)	0.900
Edema n (%)	40 (26.8)	23 (27.7)	17 (25.8)	0.789
Ascites n (%)	7 (4.7)	2 (2.4)	5 (7.6)	0.242
NIV n (%)	27 (18.1)	13 (15.7)	14 (21.2)	0.382
Cardiopathy n (%)	83 (55.7)	46 (55.4)	37 (56.1)	0.938
Ischemic heart disease n (%)	25 (16.8)	17 (20.5)	8 (12.1)	0.175
Cirrhosis n (%)	11 (7.4)	8 (9.6)	3 (4.5)	0.347
LTOT n (%)	32 (21.5)	20 (24.1)	12 (18.2)	0.383
Advanced malignancy n (%)	16 (10.7)	9 (10.8)	7 (10.6)	0.963
Confined to bed n (%)	51 (34.2)	21 (25.3)	30 (45.5)	0.010
Hospital stay (median, IQR)	10 (6–15)	10 (6–15)	9.5 (7–14)	0.875
Edema/ascites n (%)	47 (31.5)	23 (27.7)	18 (27.3)	0.953
Cardiopathy/cirrhosis/dialysis n (%)	95 (63.7)	50 (60.2)	38 (57.6)	0.742
In-hospital mortality n (%)	38 (25.5)	19 (22.9)	19 (28.8)	0.412

Data are expressed as absolute values and percentage for categorical variables and as mean ± SD or median ± interquartile range for continuous variables. p values refer to between-group comparisons (male vs. female)

CCI Charlson Comorbidity Index, SOFA Sequential Organ Failure Assessment, CRP C-reactive protein, PCT procalcitonin, NIV non-invasive ventilation, LTOT long-term oxygen therapy

The most common site of infection was the respiratory tract, accounting for 51.7% (n = 77) of cases, followed by intra-abdominal infections (17.4%, n = 26), urinary tract infections (16.1%, n = 24), skin and soft tissue infections (6.7%, n = 10) and endocarditis or catheter-related infections (2.7%, n = 4). In 5.4% (n = 8) of patients, the primary site of infection could not be determined. Bacteremia was documented in 30.2% (n = 45) of patients, and multi-drug resistant (MDR) pathogens were isolated in 12 patients (8.1%).

Baseline comorbidity burden was high, as indicated by a median CCI of 6, while the median SOFA score at admission was 3. Major comorbidity prevalence in our cohort was as follows: cardiopathy 55.7%; ischemic heart disease 16.8%; cirrhosis 7.4%; history of dialysis 0.7%. Other relevant conditions included a history of LTOT in 21.5%, advanced malignancy in 10.7% and confined to bed status in 34.2% of patients.

Regarding laboratory parameters, the median levels of CRP, PCT and lactate were 117.8 mg/l, 1.67 ng/ml and 1.77 mmol/l, respectively.

On physical examination, 26.8% of patients presented with peripheral edema and 4.7% with ascites. Additionally, 18.1% of the cohort required NIV. Hemodynamic instability was present in 6.7% of patients.

The median length of hospital stay was 10 days. In-hospital mortality was 25.5% (n = 38), including 14.7% (n = 22) within the first 7 days and 11.4% (n = 16) between days 8 and 28.

Of note, when baseline characteristics were stratified by sex, no significant differences were observed between female and male patients in terms of age, comorbidity burden, laboratory parameters or clinical presentation, except for a higher prevalence of confined to bed status among female patients (45.5% vs. 25.3%, p = 0.010).

Comparison of Baseline Characteristics Between Survivors and Non-survivors

A comparison of baseline characteristics between survivors and non-survivors highlighted several important differences (Table 2).

Table 2.

Comparison among clinical, demographic and laboratory variables between survivors and non-survivors

Variable	Survivors (n = 111)	Non-survivors (n = 38)	p value
Age, years (mean ± SD)	74.8 ± 12.75	81.3 ± 8.94	0.005
Male n (%)	64 (57.7)	19 (50)	0.412
Days from symptom onset, median (IQR)	2 (1–4)	4 (2.75–10)	< 0.001
CCI (median, IQR)	5 (4–7)	7 (6–9)	< 0.001
SOFA (median, IQR)	3 (2–5)	4 (3–6)	0.036
Hemodynamic instability % (n)	3.6% (4)	15.8% (6)	0.018
CRP, mg/l (median, IQR)	105.6 (50.9–189)	158.4 (74.7–252.4)	0.087
PCT, ng/ml (median, IQR)	1.02 (0.19–7.34)	4.12 (0.79–18.22)	0.021
Lactate, mmol/l (median, IQR)	1.6 (1.1–2.3)	2.7 (1.7–3.7)	< 0.001
Edema n (%)	33 (29.7)	7 (18.4)	0.175
Ascites n (%)	5 (4.5)	2 (5.3)	1
NIV n (%)	13 (11.7)	14 (36.8)	0.001
Cardiopathy n (%)	61 (55)	22 (57.9)	0.753
Ischemic heart disease n (%)	19 (17.1)	6 (15.8)	0.850
Cirrhosis n (%)	7 (6.3)	4 (10.5)	0.472
LTOT n (%)	19 (17.1)	13 (34.2)	0.027
Advanced malignancy n (%)	7 (6.3)	9 (23.7)	0.006
Confined to bed n (%)	28 (25.2)	23 (60.5)	< 0.001
Hospital stay (median, IQR)	11 (8–15)	5.5 (3–10.25)	< 0.001
Edema/ascites n (%)	33 (29.7)	8 (21.1)	0.300
Cardiopathy/cirrhosis/dialysis n (%)	64 (57.7)	24 (63.2)	0.552

Data are expressed as absolute values and percentage for categorical variables and as mean ± SD or median ± interquartile range for continuous variables

CCI Charlson Comorbidity Index, SOFA Sequential Organ Failure Assessment, CRP C-reactive protein, PCT procalcitonin, NIV non-invasive ventilation, LTOT long-term oxygen therapy

Non-survivors were significantly older, with a mean age of 81.3 years, compared to 74.9 years in survivors. The median time from symptom onset to ED admission was also significantly longer in non-survivors compared to survivors, suggesting that delayed presentation was associated with poorer outcomes.

The CCI was significantly higher in non-survivors than survivors, as were the presence of advanced malignancy, confined to bed status and a history of home oxygen therapy, indicating that a great burden of comorbid conditions as well as a significant functional deterioration prior to admission contributed to worsening the clinical course. The SOFA score was also higher in non-survivors than in survivors; however, the difference in median values was minimal, and the interquartile ranges were partially overlapping, suggesting a limited ability of the SOFA score to discriminate between outcome groups.

As expected, hemodynamic instability was more prevalent in non-survivors compared to survivors. Another key difference between the two groups was the use of NIV, necessary for a significantly higher proportion of non-survivor patients according to the more severe respiratory impairment in this cohort. The requirement for organ support, such as vasopressors and NIV, implies that non-survivors had more advanced organ dysfunction at baseline and needed a more aggressive initial management.

Additionally, lactate and PCT levels were higher in non-survivors, consistent with the prognostic role of inflammation and metabolic stress. Interestingly, however, CRP levels did not significantly differ between the two groups.

Instead, no significant differences were observed between survivors and non-survivors in terms of edema and ascites, even when these variables were combined. Similarly, we found no significant differences between groups regarding cardiopathy (including ischemic heart disease) or cirrhosis. Only one patient (belonging to the non-survivors) underwent dialysis, thus precluding any statistical analysis. Furthermore, when these conditions (cardiopathy, cirrhosis and dialysis) were analyzed together, no significant differences emerged between the two groups.

Lastly, the median hospital length of stay was significantly shorter in non-survivors because of a rapid clinical deterioration occurring in these patients.

No significant differences were observed between patients who died within 7 days and those who died between days 8 and 28, except for age, which was slightly higher in early deaths. This finding supports the use of all in-hospital deaths as a single outcome group in the primary analysis.

To account for the higher prevalence of confined to bed status among females, we performed an additional subgroup analysis stratified by sex. This confirmed that confined to bed status was significantly associated with in-hospital mortality in both male and female patients (χ² = 9.7; p = 0.002 and χ² = 5.7; p = 0.017, respectively), suggesting that its prognostic role was independent of sex distribution.

ROC Analysis

To evaluate the discriminative performance of the variables that differed significantly between survivors and non-survivors, we performed receiver-operating characteristic (ROC) curve analysis (Fig. 1 and Table 3). For visual clarity, only continuous variables are shown in Fig. 1.

Fig. 1.

ROC curves of selected variables for in-hospital mortality. Receiver-operating characteristic (ROC) curves of age, PCT, time from symptom onset and lactate for predicting in-hospital mortality. The blue diagonal line represents the reference (non-discriminative) line corresponding to an area under the curve (AUC) of 0.5. Cutoff values were defined using Youden’s index. Sensitivity and specificity are expressed as percentages and refer to the selected thresholds. PCT procalcitonin

Table 3.

Discriminative performance of categorical variables for in-hospital mortality based on ROC curve analysis

Variable	AUC	p value
Age ≥ 79.5 (years)	0.63	0.013
PCT ≥ 0.63 (ng/ml)	0.63	0.018
Lactate ≥ 2.21 (mmol/l)	0.71	0.000
Days from symptom onset ≥ 3.5	0.64	0.008
Hemodynamic instability	0.56	0.260
NIV	0.63	0.020
LTOT	0.58	0.120
Advanced malignancy	0.59	0.110
Confined to bed status	0.68	0.001
Organ support	0.66	0.003
Frailty composite	0.73	0.000

All variables listed are dichotomous. Some were originally categorical (e.g., NIV, LTOT), while others were derived from continuous variables using optimal cutoffs identified by Youden’s index based on ROC curve analysis. AUC values reflect the performance of each variable in predicting in-hospital mortality. Composite variables were constructed by aggregating predictors with shared biologic or clinical relevance to enhance discriminative power: Organ Support = presence of NIV or hemodynamic instability; Frailty Composite = presence of at least one among confined to bed status, LTOT or advanced malignancy

AUC area under the curve, ROC receiver-operating characteristic, PCT procalcitonin, NIV non-invasive ventilation, LTOT long-term oxygen therapy

Among the four continuous variables tested, serum lactate exhibited the highest individual discriminative ability, followed by time from symptom onset, age and PCT. Cutoff values were determined using Youden’s index and used to dichotomize the continuous variables. Once dichotomized, each variable was re-evaluated through ROC analysis to confirm that the selected thresholds preserved adequate discriminative performance (Table 3).

Regarding categorical variables, many did not achieve statistical significance (p > 0.05) and acceptable discriminative accuracy (AUC < 0.60), likely because of the small number of events. However, based on their established clinical relevance, these variables were aggregated into biologically and clinically coherent composite predictors. Specifically, a composite variable reflecting biologic vulnerability and functional exhaustion, hereafter referred to as the Frailty Composite, was derived from the presence of at least one of the following: confined to bed status, LTOT or advanced malignancy. Similarly, a second clinical composite variable, termed Organ Support, was constructed by combining the presence of NIV or hemodynamic instability as markers of organ support requirement and critical physiologic derangement. Both composite variables achieved significantly improved predictive performance (AUC = 0.73 and 0.66, respectively), as reported in Table 3.

Logistic Regression Analysis of Individual Predictors

To assess the prognostic value of the variables identified through univariate comparisons and ROC analysis, we performed independent binary logistic regression models, each adjusted for sex and age, with in-hospital mortality as the dependent outcome. All variables tested were dichotomous, either inherently or obtained by dichotomizing continuous variables. These findings are visually summarized in the forest plot (Fig. 2).

Fig. 2.

Forest plot of independent predictors of in-hospital mortality. Odds ratios (OR) and 95% confidence intervals (CI) were calculated from independent binary logistic regression models, each adjusted for sex and age, except for age ≥ 79.5 years, which was adjusted for sex only. The vertical dashed line indicates the null value (OR = 1). Frailty Composite = presence of at least one among confined to bed status, LTOT or advanced malignancy; Organ Support = presence of NIV or hemodynamic instability. PCT procalcitonin, Lac lactate, LTOT long-term oxygen therapy, NIV non-invasive ventilation

The regression models confirmed the predictive relevance of several variables, particularly elevated lactate levels, prolonged symptom duration, indicators of systemic deterioration, such as hemodynamic instability or the need for NIV, and the presence of conditions reflecting biologic vulnerability. In contrast, PCT did not reach statistical significance in the adjusted models.

Development of the ASIs Prognostic Score

To construct the Acute Severity Infection Score (ASIs), we entered all previously selected variables into a multivariable regression model, using in-hospital all-cause mortality as the dependent outcome (Table 4). Multicollinearity was assessed by calculating the variance inflation factor (VIF) for each variable, ensuring the independence of all predictors.

Table 4.

Multivariable model and weighted components of the acute severity infection score (ASIs)

Variable	Criteria	Clinical interpretation	OR (CI 95%)	B	p value	Points
Serum lactate	≥ 2.21 mmol/l	Metabolic stress	8.9 (3.1–25.4)	2.2	0.000	2
Frailty composite	≥ 1 condition	Baseline vulnerability	11.1 (3.1–39.6)	2.4	0.000	2
Organ support	≥ 1 condition	Need of organ support	3.2 (1.1–9.2)	1.2	0.027	1
Age	≥ 79.5 years	Advanced age	3.8 (1.4–10.7)	1.3	0.010	1
Days from symptom onset	≥ 3.5 days	Delayed access to care	3.7 (1.4–10.1)	1.3	0.010	1

Final multivariable logistic regression model used to construct ASIs. Each variable is presented with its dichotomization criterion, clinical interpretation, odds ratio (OR) with 95% confidence interval (CI), regression coefficient (B) and p-value. Point values were assigned based on the relative strength of each predictor as indicated by the B coefficient: variables with B ≥ 2.0 receiving 2 points, while others received 1 point. The total score ranges from 0 to 7. Frailty Composite = presence of at least one among confined to bed status, LTOT or advanced malignancy; Organ Support = presence of NIV or hemodynamic instability

LTOT long-term oxygen therapy, NIV non-invasive ventilation

The final model retained five independent predictors of mortality, selected based on statistical significance (p < 0.05), discriminative accuracy (AUC), odds ratios (ORs) with 95% confidence intervals and, most importantly, the magnitude of the regression coefficient (B), which reflected the relative weight of each variable in the model. A weighted scoring system was then derived, assigning 2 points to variables with the strongest association and 1 point to the remaining predictors. This strategy yielded a clinically interpretable score while preserving the statistical weight of each variable.

The final ASIs ranges from 0 to 7 points, with higher values indicating greater predicted risk of in-hospital mortality.

Prognostic Assessment of the ASIs and Comparison with Established Prognostic Tools

The prognostic performance of the ASIs was evaluated through ROC curve analysis (Fig. 3) and Cox regression models (Table 5) and compared with the SOFA score and CCI. The ASIs demonstrated a high discriminative ability (AUC of 0.883) and strong sensitivity and specificity at the optimal cutoff of 4.5, with CCI and SOFA scores showing lower prognostic accuracy.

Fig. 3.

Receiver-operating characteristic (ROC) curves comparing the discriminative performance of ASIs, CCI and SOFA for in-hospital mortality. ROC curves comparing the prognostic performance of the Acute Severity Infection score (ASIs), Charlson Comorbidity Index (CCI) and Sequential Organ Failure Assessment (SOFA) score in predicting in-hospital mortality. The diagonal reference line represents an AUC of 0.5, indicating no discriminative ability. Optimal cutoff values were identified using Youden’s index. AUC area under the curve, CI confidence interval

Table 5.

Prognostic value in Cox regression models

Score	Covariate adjustment	HR (95% CI)	p value	Standardized HR
ASIs	Gender	1.9 (1.6–2.3)	< 0.001	3.4
	Gender, CCI, SOFA	1.8 (1.4–2.2)	< 0.001	2.9
CCI	Gender	1.4 (1.2–1.5)	< 0.001	2.1
	Gender, ASIs	1.1 (0.9–1.3)	0.150	1.3
	Gender, SOFA	1.3 (1.2–1.5)	< 0.001	2
SOFAs	Gender, age	1.1 (1.0–1.3)	0.046	1.3
	Gender, ASIs	1.0 (0.9–1.2)	0.650	1.1
	Gender, CCI	1.1 (0.9–1.2)	0.290	1.1

Standardized HRs were calculated to compare the relative effect size of each score

HR hazard ratio, CI confidence interval, ASIs acute severity infection score, CCI Charlson Comorbidity Index, SOFA Sequential Organ Failure Assessment

In multivariate Cox regression models adjusted for gender (ASIs and CCI), or for gender and age (as age is not included in SOFA score), the ASIs demonstrated a strong and independent association with in-hospital mortality (Table 5). Specifically, each 1-point increase in the ASIs was associated with a 90% increase in the hazard of in-hospital death (HR 1.9, 95% CI 1.6–2.3; p < 0.001). To enable direct comparison among scoring systems with different scales, hazard ratios were standardized using z-score transformation. The standardized HR for ASIs was 3.4, substantially higher than that of the CCI (2.1) and the SOFA score (1.3), strengthening its superior discriminative performance.

Moreover, in fully adjusted models including ASIs, CCI and SOFA as covariates, only the ASIs retained independent prognostic significance. Conversely, both the CCI and SOFA scores lost statistical significance when adjusted for ASIs, suggesting that ASIs not only captures but also consolidates the predictive domains represented by these traditional tools.

Discussion

To date, no universally accepted diagnostic or prognostic tool for sepsis has been developed and validated, as also acknowledged in the most recent Surviving Sepsis Campaign guidelines, which emphasize the critical need for early recognition and appropriate management in the absence of a single validated prognostic score [20].

Among the tools currently available, the SOFA score is widely used to assess organ dysfunction in critically ill patients, while the CCI is employed to quantify baseline disease burden. However, neither was specifically developed to stratify mortality risk in patients hospitalized for acute infections in internal medicine settings, nor do they capture the multidimensional interplay of clinical severity, inflammatory stress and pre-existing frailty.

The SOFA score, in particular, was conceptualized and validated in ICUs to assess organ dysfunction [8–10]; thus, it lacks integration of critical prognostic dimensions such as baseline comorbidities, metabolic stress markers and clinical frailty. Moreover, its applicability in general medical wards has been questioned, as it may fail to adequately reflect the clinical complexity and prognostic heterogeneity of patients outside the ICU setting [2, 10]. Conversely, the CCI, although valuable for assessing chronic disease burden [11], has demonstrated limited predictive accuracy in patients with severe acute infections [21], as it fails to capture the acute physiologic deterioration and inflammatory response characteristic of such conditions [22]. Nevertheless, we chose to include both the SOFA and the CCI scores as comparators in our analysis, as they represent two complementary prognostic domains—acute organ dysfunction and chronic comorbidity burden—against which it was essential to benchmark a novel score specifically designed to integrate these dimensions.

Clinicians routinely integrate data from medical history, clinical examination, laboratory results and functional status to guide therapeutic decisions in patients with acute infections. Yet, no standardized instrument has been developed to objectively support this multidimensional approach.

The ASIs responds to this unmet need by incorporating the full spectrum of prognostically relevant domains: comorbidity and frailty, clinical severity at presentation and inflammatory-metabolic stress.

Each component of the ASIs represents not merely a statistical association but also a pathophysiologically grounded marker of systemic vulnerability.

Elevated serum lactate is a well-established indicator of early metabolic stress, reflecting mitochondrial dysfunction, anaerobic glycolysis and tissue hypoperfusion, even in the absence of hypotension or overt shock [23, 24]. In our cohort, we identified a threshold of 2.2 mmol/l, closely mirroring previous findings. A retrospective study in patients with sepsis reported 2.5 mmol/l as the optimal cutoff for predicting 28-day mortality [25], while an analysis of over 4000 elderly inpatients with sepsis identified a 2.4 mmol/l threshold as significantly associated with short-term mortality [26].

Another key item in the ASIs is the delay between symptom onset and hospital admission. While current guidelines emphasize early in-hospital antibiotic administration, emerging evidence suggests that pre-hospital delay may be equally critical, particularly in older or functionally impaired patients [27, 28]. Several large-scale studies have demonstrated that delays of around 4 days, and in some cohorts as little as 24 h, are significantly associated with higher mortality among patients with acute infections or sepsis [29, 30]. In this context, the inclusion of a symptom-to-admission interval > 3.5 days yields a relevant time-sensitive prognostic signal, reflecting an inflection point beyond which unrecognized infection may evolve toward systemic decompensation.

Equally important, although often considered a background risk factor, advanced age profoundly influences outcomes in acute infection. Immunosenescence, diminished physiologic reserve and multimorbidity contribute to increased vulnerability, even in the absence of overt organ failure. In addition, atypical presentations and delayed recognition may further compromise prognosis in older adults [31]. In our cohort, 79.5 years emerged as the optimal threshold for mortality risk stratification, consistent with previous evidence [32, 33].

Of particular significance is the composite variable encompassing hemodynamic instability or the need for NIV. The former is a well-established predictor of mortality in septic patients with extensive evidence linking hypotension and vasopressor use to adverse outcomes [34–36]. The need for NIV, similarly, reflects early respiratory compromise. In a large real-world cohort, its use at admission was significantly more common among non-survivors, independently of success or failure [37]. The presence of either event marks the point at which systemic homeostasis has failed and organ support becomes necessary, a clinical threshold beyond which prognosis deteriorates significantly.

Finally, the frailty composite offers a dynamic measure of biologic reserve, capturing the functional exhaustion that static comorbidity scores often miss. In our cohort, it showed the strongest prognostic role among all ASIs components. This is in line with previous studies reporting a nearly fourfold rise in death among confined to bed individuals admitted for bloodstream infections [38] and 1-year mortality > 60% in functionally dependent geriatric inpatients [39]. Similarly, LTOT was associated with markedly higher 30-day and 1-year mortality in patients with COPD and interstitial lung disease, even after adjustment for disease severity [40, 41]. Obviously, advanced malignancy remains one of the strongest and most consistent predictors of poor prognosis in infection. Large-scale data show infection-related mortality rates nearly three times higher than in the general population [42] and up to 70% in oncologic patients with untreated bloodstream infections [43, 44].

Altogether, the items we considered in this study show biologic plausibility and a strong alignment with real-world evidence. By integrating each domain into a single, pragmatic score, the ASIs may provide clinicians with a powerful tool to stratify patients at the time of hospital admission. Notably, it was developed and validated in a real-world internal medicine cohort, enhancing its applicability across various care settings. Furthermore, our cohort was not preselected based on a diagnosis of sepsis, unlike most existing studies, thereby avoiding the methodologic bias of validating a prognostic tool within a population defined by the very score it seeks to replace.

Unlike tools dependent on complex technologies or costly biomarkers, the ASIs is built entirely from variables readily available in standard clinical practice. It does not rely on advanced imaging, specialized assays or resource-intensive infrastructure. Instead, it synthesizes objective parameters that any trained clinician can assess, regardless of geographic location or institutional resources. This universality enhances the score’s potential for widespread adoption, particularly in resource-limited settings.

The performance of the ASIs in this study was striking. It demonstrated the highest discriminative accuracy among all tools evaluated. It also showed the strongest predictive performance for in-hospital mortality and maintained independent prognostic significance in multivariate Cox regression models adjusted for both SOFA and CCI. In contrast, both SOFA and CCI lost statistical significance when adjusted for ASIs, suggesting that ASIs encapsulates and expands upon the predictive domains captured by these scores. Beyond SOFA and CCI, other early warning scores, such as MEWS, NEWS, NEWS2 or EWS, are largely used. These tools, primarily designed for the rapid detection of acute deterioration in hospitalized patients, rely almost exclusively on dynamic changes in vital signs and are therefore valuable for real-time monitoring and for triggering escalation of care in patients with acute infections. In contrast, the ASIs was designed to integrate multiple domains of patient health, such as laboratory biomarkers, chronic comorbidities and clinical status, offering a more complete assessment of mortality risk and complementing existing early warning systems to support individualized clinical decision-making in medical wards.

Our findings support the hypothesis at the heart of this project: that the early identification of high-risk infections is, in essence, the most effective strategy for improving sepsis management. On the other hand, what is sepsis if not an infection with high mortality? Rather than waiting for fulminant organ dysfunction to manifest, the ASIs aims to anticipate it by identifying, early and objectively, those infections most likely to progress toward adverse outcomes. In doing so, it shifts the diagnostic approach from reactive to proactive. Moreover, the ASIs reflects how experienced physicians already reason in clinical practice. It translates the multidimensional assessment process—comorbidities, frailty, hemodynamic and respiratory compromise, inflammatory and metabolic stress—into a reproducible, evidence-based tool. In this sense, the ASIs would not be a replacement for clinical judgment but rather its codification and amplification.

This study is not without limitations. Its retrospective nature may be subject to selection bias and confounding, although we mitigated these risks through rigorous inclusion criteria, clearly defined operational variables and consistent data extraction within the first 12 h of hospital admission. Implicitly, the retrospective design of the study did not allow to systematically measure additional prognostic scores, including MEWS, NEWS, etc. [20]. In this regard, prospective validation of ASIs across larger and more diverse populations, along with its comparison to other infection prognostic scores, is already planned.

Conclusion

The ASIs may represent an innovative, multidimensional and clinically grounded tool for the early risk stratification of patients hospitalized with acute infections. By integrating easily accessible biomarkers, it may be universally adopted for real-world clinical decision-making well before sepsis becomes fully manifest.

Author Contribution

Alessio Comitangelo and Alfredo Vozza contributed equally to this work and share first authorship. They were responsible for the concept and design of the study, acquisition and interpretation of data, statistical analysis and drafting of the manuscript. Giovanna Ditaranto, Giuseppe Re, Ada Berloco, Erasmo Porfido, Domenico Comitangelo, and Sara Madaghiele contributed to the acquisition of data and critical revision of the manuscript for important intellectual content. Carlo Custodero and Andrea Portacci contributed to the development of the methodology, statistical analysis, and interpretation of data. Cosimo Tortorella and Giuseppina Piazzolla contributed equally to the supervision of the project, interpretation of the data and critical revision of the manuscript. All authors reviewed and approved the final version of the manuscript.

Funding

This study received no external funding. The journal’s publication fee is partially supported by the Interdisciplinary Department of Medicine, University of Bari Aldo Moro. The remaining costs will be covered by the authors.

Data Availability

The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Conflict of Interest

Alessio Comitangelo, Alfredo Vozza, Giovanna Ditaranto, Giuseppe Re, Ada Berloco, Erasmo Porfido, Carlo Custodero, Domenico Comitangelo, Sara Madaghiele, Andrea Portacci, Cosimo Tortorella, and Giuseppina Piazzolla declare that they have no conflicts of interest.

Ethical Approval

The study was conducted in accordance with the general ethical principles for medical research involving human subjects of the Declaration of Helsinki [19]. The study protocol was formally approved by the Clinical Investigation Ethics Committee of the University of Bari (ID: Studio ASIs 2025, protocol number 2150/CEL, version 1.01, 19 March 2025). As a retrospective study based on anonymized clinical data, formal informed consent was waived by the ethics committee.