Sepsis is among the major global health concerns, in terms of morbidity, mortality and economic burden (1). It results from a dysregulated host response to infection, generating persistent, harmful inflammation with organ dysfunction and clinical deterioration, eventually progressing to refractory hypoperfusion, critical multiorgan failure, and death in a significant subset of patients (2).

The incidence of sepsis has increased in recent decades, in parallel with the increase in aging, chronic diseases and antimicrobial resistance (3–5). In fact, sepsis represents the main cause of in-hospital mortality, contributing directly or indirectly to at least half of all in-hospital deaths, as well as being responsible for long and complicated hospital stays, with frequent readmissions (5–7).

Sepsis is a highly heterogeneous and time-dependent life-threatening condition, thereby early identification of patients developing sepsis and at risk of progression is pivotal to optimize treatment strategies, prevent worse outcomes and reduce mortality (8). Most of the literature regarding the epidemiology, management, and prognosis of sepsis to date has focused on patients in emergency departments and intensive care units (ICU) (9, 10); however, owing to the aging of the general population and limited ICU resources, sepsis is increasingly managed in medical departments, where at least two-thirds of these patients are treated (7, 8, 11–16), but clinical evidence regarding the effectiveness of guidelines and bundles in this hospital setting is still largely lacking, and little is known about the characteristics and outcomes of patients with sepsis admitted to general medicine wards.

In this observational study, we investigated the clinical course of patients with sepsis admitted to four internal medicine units in Lombardy (Italy) over the last decade, focusing on key hospital clinical outcomes, such as length of stay (LOS), ward-to-ICU transfer and in-hospital mortality rates. Additionally, we stratified patients based on blood culture results, thereby looking for potential differences in clinical course between culture-positive versus culture-negative, polymicrobial versus monomicrobial, and multidrug-resistant (MDR) versus non-MDR sepsis patients, and we attempted to identify blood-based microbiological and laboratory biomarkers that could serve as potential predictors of distinct types of sepsis and outcomes. Finally, we studied the potential impact of the COVID-19 pandemic on sepsis and the trend of antimicrobial resistance in the latest years.

This study provides novel insights into the epidemiological trends and clinical outcomes of sepsis in internal medicine units, an underexplored hospital setting, over a 12-year period. This work is also among the first to comprehensively analyze the prognostic significance of several laboratory variables in this environment, identifying key biomarkers, such as ferritin, that may predict adverse outcomes. These findings bridge a critical gap in sepsis management outside the ICU, where limited resources and an aging patient population represent unique challenges that necessitate improved diagnostic and therapeutic strategies.

Univariate analyses revealed discrete associations between sepsis types and hospital outcomes, in particular: culture-negative sepsis was associated with shorter stay and higher mortality (more pronounced hypoperfusion with acute kidney injury and rapid progression); polymicrobial sepsis was associated with higher ICU transfer rates (more pronounced hypoventilation with acute respiratory distress syndrome and higher need of oxygen and ventilation); while MDR⁺ and/or polymicrobial sepsis were associated with longer stay and higher mortality (complicated course).

Multivariate analyses confirmed reciprocal associations of culture-positive sepsis with prolonged hospitalization and oxygen requirement, specifically: MDR⁺ cultures predicted polymicrobial sepsis and the latter predicted longer hospital stay; in turn, a longer hospital stay predicted culture-positive sepsis, and higher oxygen need predicted culture-positive and polymicrobial sepsis. In accordance with our results, a recent meta-analysis showed that patients with culture-positive sepsis had longer duration of mechanical ventilation and longer hospital stay compared to culture-negative ones (21), while other authors reported that patients with polymicrobial sepsis had more need of ventilation and longer hospital stay compared to monomicrobial cases (15, 22).

Notably, we observed that hospitalizations with the highest LOS peaks precede those with the highest mortality peaks by 1 year. Hospitalizations and antibiotic therapies in previous months have been independently associated with occurrance of MDR⁺ septicemia at readmission (23, 24), and our data confirmed that polymicrobial sepsis often harbors MDR⁺ bacteria (e.g., VRE, carbapanem-resistant Klebsiella and other Gram-negative bacteria), which may ultimately be responsible for higher mortality (15, 25). In this regard, increased circulating levels of ferritin were herein found to predict both MDR⁺ sepsis and in-hospital mortality.

While on the one hand we have recorded an overall reduction in MDR⁺ sepsis (and in-hospital mortality) in recent years, due to a significant drop in the resistance of Gram-negative bacteria to cephalosporins, fluoroquinolones and aminoglycosides, with stability or at least numerical reduction of discrete resistant Gram-positive strains (e.g., MRSA), on the other hand we have highlighted that septicemia caused by carbapenemase-producing (e.g., KPC and others) and ESBL⁺ Gram-negative bacteria, as well as VRE, are on the rise, in agreement with other reports (24, 26, 27). The inappropriate use of broad-spectrum nosocomial antimicrobials, such as carbapenems, protected ureidopenicillins (e.g., piperacillin/tazobactam), and glycopeptides (e.g., vancomycin), to treat presumed bacterial coinfections during the COVID-19 pandemic (28) likely contributed to increase the resistance to these specific antibiotics. Moreover, the rise in polymicrobial sepsis herein observed in post-COVID years may reflect the immunosuppressive effects of SARS-CoV-2 among elderly patients (29). Nonetheless, the overall decrease in MDR⁺ rates seemed to result in a numerical reduction of sepsis mortality over the period 2016–2023, a trend that became fully significant when considering the entire observation period 2012–2023, in line with previous literature and meta-analyses (11, 30). In any case, there remains a need for further implementation of antibiotic stewardship programs in internal medicine units, with less protracted and more targeted therapies, in the aim to be more effective against sepsis but less harmful on antimicrobial resistance (7, 27, 31).

Another unmet need in the management of sepsis in medical wards is the lack of validated biomarkers of MDR⁺ sepsis and poor prognosis. Here, we identified circulating predictors of distinct types of sepsis and hospital outcomes, highighting differences between commonly used laboratory variables, such as leukocyte and platelet counts, CRP, procalcitonin, lactate, and ferritin levels. In particular, CRP elevation predicted rapidly evolving culture-negative sepsis, whereas decreased leukocyte counts predicted longer hospital stay; in accord with previous observations (32), lactate elevation predicted ICU transfer; and, most remarkably, ferritin elevation along with increased leukocyte counts were predictive of MDR⁺ sepsis, while further ferritin elevation along with decreased platelet counts were highly predictive of in-hospital mortality.

In our study, baseline procalcitonin was higher in culture-negative sepsis but, in agreement with other reports (8, 33), did not emerge as an independent predictor. Previous studies have highlighted that procalcitonin is useful outside the ICU for early recognition of sepsis, including culture-negative ones, from other inflammatory syndromes, while other indices (e.g., MR-proadrenomedullin) are instead more sensitive biomarkers of culture-positive forms; hence, baseline procalcitonin has an established diagnostic role in early sepsis (34, 35). Nevertheless, since procalcitonin level rapidly declines if antibiotic treatment is effective, its change within few days from admission (delta-procalcitonin) may have some prognostic role in sepsis patients admitted to internal medicine wards (33). In either case, serum procalcitonin levels are closely related to the presence or load of a bacterial pathogen. By contrast, serum ferritin levels are associated with the magnitude of the host inflammatory response, in particular macrophage activation (36, 37): in case the pathogen is resistant to treatment (e.g., first antibiotics administered in the emergency room), leukocytes such as macrophages and cytotoxic T cells will be aberrantly activated in the attempt to kill and clear it; if this response is protracted over time (e.g., in case of MDR⁺ pathogens), it may lead to macrophage activation syndrome-like features including hyperinflammation uncoupled to the pathogen load, macrophage dysregulation/reprogramming and hemophagocytosis, which ultimately account for thrombocytopenia and rapid progression to death (37–40). Recently, Fang et al. (41) observed that serum ferritin ≥591.5 ng/mL acts as an independent predictor of in-hospital mortality in ICU patients (OR 2.29, 95% CI 1.83 to 2.87, p < 0.001), with moderate predictive power (AUC = 0.651). Here we show that, in non-ICU patients, baseline ferritin >553 ng/mL is a much stronger independent predictor of in-hospital mortality (OR 18.37, 95% CI 3.095 to 188.4, p = 0.0043), with high predictive power (AUC = 0.866), thus suggesting its use as a key prognostic biomarker of sepsis in medical departments, even better than more traditional indices such as baseline lactate or procalcitonin.

The literature regarding culture-negative sepsis is still limited. Culture-negative sepsis is probably an underestimated entity, since patients without an identified bloodstream pathogen are less likely to obtain a sepsis diagnosis (42). Indeed, blood culture sampling has been reported to identify the causative organism of sepsis only in 15–30% of the patients (42, 43) and, accordingly, proportion of patients diagnosed with culture-positive sepsis in our series was only 24.9%. Culture-negative sepsis would represent at least half of total cases and its proportion is on the rise (16, 42, 44) but, despite its great prevalence, it is largely understudied and prognosis is somewhat controversial (42, 45). Our observation of higher in-hospital mortality in culture-negative sepsis aligns with previous reports by Gupta et al. (44), who noted that treatment remains empiric and often less effective in such cases. However, the debate continues as Mellhammar et al. (42) observed lower 90-day mortality in culture-negative cases, suggesting the need for further studies. On the other hand, it was recognized that culture-negative sepsis is more frequent in case of abdominal or pulmonary infections, rather than urinary ones (42), and the former actually show higher mortality than the latter (46, 47).

The rate of in-hospital mortality from sepsis in our experience was 26%, in line with global studies (3, 4, 48). Sepsis mortality changes depending on organ dysfunction: less than 20% in cases without organ dysfunction [“sepsis” according to Sepsis-1/2 criteria (18, 19)], between 20 and 50% in cases with organ dysfunction [“sepsis” according to Sepsis-3 criteria (2) or “severe sepsis” according to Sepsis-1/2 criteria], and over 50% in patients with septic shock (3). Fleischmann et al. (4) estimated that hospital mortality was 17% for “sepsis” and 26% for “severe sepsis” during the decade 2006–2015, whereas an updated and extended systematic review up to 2019 and meta-analysis of total patients with “sepsis,” defined according to clinical criteria (Sepsis-1/2/3) or relevant ICD codes, concluded that 26.7% of hospital-treated patients vs. 49.1% of ICU-treated patients died prior to discharge (48).

Our results also show several similarities with those reported in recent Italian studies focused on sepsis in medical wards (8, 16). In the SEMINA study, in particular, the composite outcome “in-hospital mortality or ICU transfer” was achieved in around 37% of sepsis patients (31.2% death + 5.3% ICU), as in our series (26% death + 11% ICU); furthermore, in accord with our observations, patients with culture-negative sepsis were the majority, with at least numerically higher mortality rates; and non-survivors had a shorter stay, most often less than 2 weeks, whereas a longer stay is highlighted here to significantly protect against in-hospital mortality (16). Finally, the subsets of patients who achieved the poorest outcomes had higher FiO₂ requirements and lower PaO₂/FiO₂ ratios, higher creatinine levels, lower platelet counts, and higher frequencies of MDR⁺ cultures (8, 16).

Our work has several strengths. The first is certainly the large number of patients included in the study. The literature on the clinical course and prognosis of sepsis in medical wards is still scarse. Using the innovative i2b2 computational methodology, we were able to collect and analyze real-life data from up to 4,375 sepsis patients admitted specifically to internal medicine units. Second, we stratified patients based on blood culture findings, thus identifying interesting associations between microbiological and clinical characteristics, as well as useful circulating biomarkers predicting sepsis types and outcomes. Third, we studied the most recent epidemiological trends over broad time frames, examining key hospital outcomes, antimicrobial resistance and the potential impact of the COVID-19 pandemic on sepsis presentation and severity.

This study also has some limitations, including retrospective design and inclusion of patients according to ICD codes rather than Sepsis-3 or other clinical criteria (8, 16). Future studies should adopt standardised diagnostic criteria to ensure consistency. On the other hand, most of the studies conducted so far on sepsis in non-ICU settings were retrospective and based on ICD-coded hospital discharge databases, and no significant differences in mortality were found between sepsis patients identified based on ICD-codes vs. clinical criteria, nor based on Sepsis-1 vs. Sepsis-3 criteria (48). According to the third international consensus on sepsis and septic shock (Sepsis-3, 2006) (2), patients with a quick Sequential Organ Failure Assessment (qSOFA) ≥2 should be surveilled as at higher risk of in-hospital death. However, the most recent international guidelines (Survival Sepsis Campaign, 2021) (9) recommended against using qSOFA as a single screening tool for sepsis or septic shock compared to SIRS, National Early Warning Score (NEWS) or Modified Early Warning Score (MEWS), due to higher specificity but lower sensitivity in early identification of sepsis patients and prediction of sepsis outcomes, while the diagnosis of sepsis remains confirmed in case of a SOFA score ≥2. Although our analysis did not include clinimetric assessment of mental status, respiratory rate, heart rate, blood pressure, urine output and use of vasopressors for deriving SIRS, qSOFA or septic shock criteria (8, 14–16, 49), we did include all laboratory variables considered in the SOFA score. Other limitations of this study are lack of specific data about hospital-acquired sepsis (16, 50) and lack of statistical adjustments for confounding variables such as age, sex, comorbidities, sites of primary infection, number of blood culture sets, and timing of antibiotic therapy (42, 46, 47). In fact, comorbidities might importantly affect hospital outcomes as well as circulating levels of biomarkers.

In conclusion, despite some limitations, our study sheds new light on the clinical, epidemiological, microbiological and prognostic features of sepsis patients managed in internal medicine units. High-risk conditions such as culture-negative sepsis, polymicrobial sepsis and septicemia by ESBL⁺, carbapenemase-producers and VRE are on the rise, highlighting the need for new management strategies. Results suggest incorporating ferritin as a routine biomarker in risk stratification for sepsis patients in medical wards, where targeted interventions could be developed for high-ferritin patients to ameliorate their outcomes. Further prospective, multicenter studies are needed to validate the role of ferritin and other biomarkers in predicting outcomes and guiding treatment.