Intensive care unit-based mortality risk model construction for severe fever with thrombocytopenia syndrome patients: a retrospective study

Abstract

Objective

This study develops a predictive model to evaluate mortality risk in severe fever with thrombocytopenia syndrome (SFTS) patients in intensive care units (ICU) to improve the accuracy of prognosis and guide the optimization of treatment strategies.

Methods

In this study, a retrospective analysis was conducted on severe SFTS patients admitted to the ICU between July 2019 and October 2023. Patients were categorized into survival and mortality groups. Multivariate logistic regression was performed to determine independent risk factors (IRFs) for mortality. In addition, the nomogram model was constructed and its performance was assessed through ROC curves.

Results

The study comprised 218 severe SFTS patients. The mortality group showed significantly lower Glasgow Coma Scale (GCS) scores, oxygenation indices, and higher levels of several serological markers, log10(virus loads), and lactic acid. Multivariate analysis identified GCS score [odds ratio (OR) = 0.66, P < 0.001], log10(virus loads) [OR = 2.24, P = 0.001], lactic acid [OR = 1.60, P = 0.01], and cystatin C [OR = 1.80, P = 0.049] as IRFs for mortality. A nomogram incorporating these IRFs demonstrated excellent predictive accuracy (AUC = 0.92, 95% CI: 0.88–0.96), with a sensitivity of 76% and a specificity of 91%. This model showed adequate fit and good clinical applicability.

Conclusion

The nomogram model, based on GCS score, log10(virus loads), lactic acid, and cystatin C, offers clinical utility in predicting 28-day mortality for severe SFTS patients, though further validation is needed.

Introduction

Severe fever with thrombocytopenia syndrome (SFTS) is an acute zoonotic disease caused by Dabie bandavirus (DBV) infection [1]. First identified in 2009 when Chinese researchers isolated the virus from a patient’s blood sample in Henan Province [2], SFTS has since been responsible for recurrent outbreaks. Between 2011 and 2020, Shandong Province experienced substantial outbreaks, accompanied by an average annual incidence of 0.46 per 100,000 [3]. Although the national mortality rate in China declined from 10.58% in 2011 to 5.07% in 2021 [4], severe cases remain associated with high fatality, reaching 44.7% in patients with complications such as multiple organ failure and central nervous system involvement [5]. Critical manifestations include respiratory failure, altered consciousness, skin ecchymosis, gastrointestinal bleeding, and pulmonary hemorrhage, often progressing to multiple organ failure and fatal outcomes. Despite advancements in clinical management, mortality among severe SFTS patients remains high, and the underlying pathophysiology remains incompletely elucidated. This study conducts a retrospective analysis of severe SFTS cases admitted to intensive care units (ICU), identifies risk factors influencing 28-day mortality, and constructs a predictive model for mortality assessment. The findings contribute to the early recognition of high-risk patients and support prompt intervention strategies to improve survival outcomes in severe SFTS cases.

Materials and methods

Research subject

This retrospective study included patients with severe SFTS in the ICU between July 2019 and October 2023. Based on their 28-day outcomes, subjects were divided into survival and death groups (Fig. 1).

Fig. 1

Flow diagram of study profile

Inclusion criteria

① Detection of novel Bunyavirus nucleic acid in blood samples, verified via reverse transcription polymerase chain reaction (RT-PCR) in the laboratory [6]. ② All cases met the criteria for severe and critical classifications as outlined in the Consensus on Diagnosis and Treatment of Severe Fever with Thrombocytopenia Syndrome (2022 Edition).

Exclusion criteria

① Clinically diagnosed cases without pathogenic confirmation; ② Patients with co-infections of other viruses; ③ Cases with incomplete data.

Ethics

This study had been approved by the Ethics Committee of adhered to medical ethical standards and received approval from the hospital’s ethics committee. The need for written informed consent was waived.

Data collection

Demographic data

The demographic data collected comprised age, gender, and the interval between illness onset and ICU admission. Within 24 h of admission, the most severe Glasgow Coma Scale (GCS) score prior to sedation, Acute Physiology and Chronic Health Evaluation (APACHE-II) score, and the presence of mental status alterations, bleeding events, respiratory infections, and bloodstream infections were recorded. Additional data included whether patients received glucocorticoid or antiviral treatment (favipiravir or ribavirin), mechanical ventilation, and continuous renal replacement therapy (CRRT) in hospitalization. The primary outcome measure was 28-day mortality post-ICU admission.

Laboratory indicators

Laboratory indicators assessed within 24 h of ICU admission included log10 (virus loads), lactic acid, oxygenation index, fasting blood glucose (BS), platelets (PLT), activated partial thromboplastin time (APTT), D-dimer (DD), alanine aminotransferase (ALT), aspartate aminotransferase (AST), blood urea nitrogen (BUN), creatinine (Cr), cystatin C (CYS-C), high-sensitivity troponin, creatine kinase, lactate dehydrogenase (LDH), procalcitonin (PCT), and high-sensitivity C-reactive protein (hs-CRP), among others.

Statistical methods

Data analysis was carried out using SPSS 26.0 and R 3.5.3 software. Results without normal distribution were reported as median (interquartile range) [M (Q_L, Q_U)] and discussed with Mann-Whitney U tests. Categorical variables were described as frequencies or percentages and explored via χ² tests. Logistic regression was conducted to determine independent risk factors (IRFs) for mortality in severe SFTS patients, and a nomogram was developed. The calibration curve assessed the agreement between predicted and actual mortality risk (MR) probabilities. A receiver operating characteristic (ROC) curve was plotted to evaluate the predictive accuracy for mortality, while decision curve analysis (DCA) was utilized examined its clinical utility in forecasting mortality in severe SFTS patients.

Results

Comparison of clinical data between surviving and non-surviving severe SFTS patients

The study involved 218 patients diagnosed with severe SFTS, including 105 individuals in the survival group and 113 in the non-survival group. There were no significant differences between the groups for age, gender, APACHE-II score, neutrophil (NEU) and lymphocyte (LYM) counts, presence of bloodstream infection, or administration of antiviral therapy (P > 0.05). The non-survival group showed a significantly shorter hospital stay. It was clear that there was almost no difference in time from onset to ICU admission between the two groups. By comparison with the survival group, the non-survival group displayed significantly higher levels of ALT, AST, BUN, Cr, CYS-C, and HsCRP. Furthermore, the non-survival group demonstrated substantial reductions in GCS score and oxygenation index, accompanied by elevated lactic acid levels. This group also had a higher proportion of patients with altered consciousness, respiratory infections, and bleeding manifestations, as well as a greater number of individuals receiving glucocorticoid therapy, mechanical ventilation, and CRRT (Table 1).

Table 1 Comparison of demographic characteristics, baseline features, and biochemical indicators between surviving and non-surviving severe SFTS patients in ICU

Univariate analysis and multivariate logistic regression analysis

As can be seen from the univariate analysis, there were significant differences between survival and non-survival groups in terms of GCS score, log10(virus loads), lactic acid levels, oxygenation index, AST, BS, albumin (ALB), CysC, DD, and APTT (P < 0.05). Little significant differences can be noticed for other variables (P > 0.05).

In accordance with the multivariate logistic regression analysis evaluating MR among severe SFTS patients, the ten significant risk factors in the univariate analysis (P < 0.05) were defined as the independent variables. Results indicated that GCS score, log10(virus loads), lactic acid, and CYS-C were IRFs for 28-day mortality in severe SFTS patients (all P < 0.05) (Table 2).

Table 2 IRFs for predicting poor clinical outcomes obtained from univariate and multivariate logistic regression analyses

Development of a nomogram model for predicting MR in severe SFTS patients

A nomogram prediction model was constructed based on four IRFs identified through multivariate logistic regression analysis (Fig. 2). Each variable’s score corresponded to its placement on the scale. The aggregate scores of all variables were mapped onto a total score scale, which was used to assess the MR in severe SFTS patients. This model enabled the prediction of the 28-day MR probability for individuals with severe SFTS.

Fig. 2

Nomogram model for predicting MR in SFTS patients

Verification of the nomogram for predicting MR

Calibration curve analysis

Internal verification of the nomogram model was carried out according to the Bootstrap method with 1,000 resamples. A calibration curve (Supplementary Fig. S1) was generated to assess the model’s calibration. The Hosmer-Lemeshow goodness-of-fit test indicated the satisfactory fit (P > 0.05), confirming a high degree of model calibration.

ROC curve analysis

The nomogram’s area under the curve (AUC) for predicting MR in severe SFTS patients was 0.92 [95% confidence interval (95% CI): 0.88–0.96], with a sensitivity of 76%, a accuracy of 85% and a specificity of 91% (Fig. 3; Table 3). These results indicate that the nomogram demonstrates good predictive accuracy.

Fig. 3

AUC of the nomogram for mortality prediction in SFTS patients

Table 3 Predictive values of nomograms constructed in all patients and in different subgroups for adverse clinical outcome

DCA analysis

The DCA indicated considerable net benefit values across a risk threshold probability of 1–90%, confirming the nomogram model’s strong clinical applicability (Supplementary Fig. S2).

Discussion

The investigation revealed a 51.83% mortality rate among severe SFTS patients in the ICU. GCS score, log10(virus loads), lactic acid, and CYS-C were identified as IRFs for 28-day mortality in this cohort. A nomogram, developed using these risk factors, improved prognostic accuracy for severe SFTS patients and may offer valuable clinical application.

The pathogenesis of SFTS remains poorly understood, and there is no specific treatment. In this study, a significantly higher mortality rate was observed among critically ill SFTS patients in the ICU, exceeding the 44.7% reported by Cui N et al. [4]. SFTS is classified into mild, moderate, severe, and critical stages, based on symptom severity and organ involvement [7]. Severe and critical cases, in particular, are characterized by rapid disease progression and frequent complications, including multiple organ failure. These patients often require interventions such as mechanical ventilation, CRRT, glucocorticoid therapy, and plasma exchange, making them the primary focus of treatment efforts. Kim et al. reported that combined treatment with intravenous immunoglobulin (IVIG) and other drugs yielded positive therapeutic outcomes in severe cases [8]. Song X et al. demonstrated that critically ill SFTS patients treated with a combination of TPE (Therapeutic Plasma Exchange) and ribavirin showed improvements in both clinical and laboratory parameters [9]. However, many practical problems appeared in the treatment process of real-world scenarios. For example: 1). The optimal starting time for organ support and drug treatment measures still needs further research; 2). Blood resources are relatively scarce; and 3). Some patients face financial difficulties and may not be able to afford the cost of some expensive drugs since SFTS mainly occurs in rural areas. Compared with previous studies [10, 11], this model was targeted at severe SFTS patients in the ICU who had a high mortality rate. Simple and actionable predictors within the ICU were used to early identify high - risk ICU patients. For patients with more severe conditions and poorer prognoses, physicians may consider more aggressive treatment measures, such as early intervention, activation of advanced life support equipment, careful selection of the treatment timing for glucocorticoids, immunoglobulin, plasma exchange, etc., and rational allocation of scarce and expensive resources. For patients with relatively milder conditions, the treatment intensity can be appropriately adjusted to avoid overtreatment and reduce the waste of medical resources. Further studies will be carried out on the therapeutic effects of various treatment methods, hoping to provide some references for the rescue and treatment of critically ill SFTS patients in the future.

SFTS patients often present with neurological symptoms, commonly classified as SFTS-associated encephalopathy/encephalitis [12]. These symptoms typically include headaches, seizures, cognitive disturbances, irritability, limb convulsions, cognitive deficits, and altered consciousness [13,14,15]. Central nervous system involvement may result either directly from the virus’s neurotropic nature or indirectly through a cascade of neurological effects triggered by elevated cytokine levels [4, 16]. A statistical analysis of clinical manifestations in 59 SFTS patients by GAI et al. [17] revealed that 90.0% of the non-survivor cohort exhibited prominent neurological symptoms, including apathy, drowsiness, coma, muscle tremors, and convulsions, while fewer than 37.0% of the survivors presented with these symptoms. This demonstrates a high association between neurological symptoms and fatal outcomes in SFTS patients, emphasizing the need for early diagnosis and intervention to potentially reduce mortality. Several subsequent studies [18,19,20] have further supported this relationship. Additionally, some research [21, 22] has incorporated neurological symptom scoring into predictive models, demonstrating that the GCS score is a robust predictor of poor prognosis in SFTS patients. The current study identifies the GCS score within 24 h of ICU admission as a reliable IRF for mortality in severe SFTS cases. This metric is accessible, rapid, and readily available for clinicians. Lower GCS scores correlate with a higher risk of mortality and adverse prognosis, reinforcing its utility in predicting poor outcomes.

The study results revealed that, compared to the survival group, the viral load in the non-survival group was significantly higher, establishing SFTS viral load as an IRF for mortality in critically ill SFTS patients in the ICU. This aligns with the findings of Li J et al. [23], whose research demonstrated a correlation between SFTSV viremia levels and disease severity. Similarly, Zhang YZ et al. [24] reported that a high viral RNA level at admission was associated with fatal outcomes, suggesting that elevated viral replication acts a vital part in the pathogenesis of the disease.

Current work revealed vastly elevated lactic acid levels in the non-surviving cohort by comparison with those in the surviving cohort. Among these patients, 61.93% required invasive mechanical ventilation, 21.1% underwent continuous renal replacement therapy (CRRT), 71.56% developed respiratory tract infections, and 16.51% experienced bloodstream infections. Multiple organ dysfunction and infections from other pathogens contribute to inadequate tissue perfusion and metabolic disturbances. Elevated lactate levels often indicate impaired cardiovascular function, tissue hypoxia due to insufficient perfusion, and enhanced lactate production [25]. Higher lactate concentrations are typically observed in more critically ill patients [26]. This study identified lactate as an IRF for mortality in severe SFTS patients, aligning with the findings of Hao C et al. [27]. Baysan M et al. demonstrated that 24-hour lactate levels at admission were valuable predictors of in-hospital mortality in critically ill septic patients [28]. The role of lactate in sepsis prognosis and risk stratification, particularly in critically ill patients, is widely acknowledged [29]. Lactate measurements, easily performed using point-of-care testing (POCT) devices in ICUs, provide critical insights.

Sepsis is the leading cause of acute kidney injury (AKI) in ICU settings. Due to the nature of renal reserve and AKI dynamics [30], changes in serum creatinine levels often manifest with a delay. Serum CYS-C, with its smaller molecular weight and stable expression in the bloodstream, is exclusively cleared by the kidneys, making it a reliable endogenous marker for glomerular filtration [31]. However, the relationship between AKI and SFTS remains poorly understood. Studies have shown that AKI is an IRF for mortality in SFTS patients [32], particularly in those with stage 2 or 3 AKI, who typically experience higher mortality rates [33]. Jiao et al. [34] found that CYS-C levels were significantly elevated in severe and fatal SFTS cases compared to mild and non-fatal ones (P < 0.05). CYS-C was identified as an IRF for both disease severity and mortality, with a CYS-C level ≥ 1.23 mg/L corresponding to a 5.487-fold higher risk of death compared to levels < 1.23 mg/L. In this work, CYS-C levels in the non-surviving group were significantly higher than that of the surviving group and were confirmed as an IRF for mortality. This suggests that CYS-C may be a more sensitive marker of AKI in critically ill SFTS patients in the ICU than alternative biomarkers.

The nomogram developed to predict MR in severe SFTS patients highlights that lower GCS scores, along with elevated log10(virus loads), lactic acid, and CYS-C levels, significantly increase the total score. Clinically, the model incorporates indicators that are readily available through routine clinical data or existing scoring systems, ensuring ease of application. According to the statistical analysis, the AUC of the LINE graph model reached 0.92, demonstrating strong discriminative ability. The Hosmer-Lemeshow goodness-of-fit test further verified the adequacy of this model. Therefore, the MR prediction model offers considerable clinical relevance and accuracy.

There are some limitations. As a retrospective study, potential biases in data collection and analysis exist, compounded by a relatively small sample size. Although the GCS score can be influenced by sedation, we resolved this by obtaining the score before sedation. Due to limitations in research resources and time, we were unable to follow up all patients for as long as 90 days. Additionally, due to the slow recovery of some critically ill patients, after being treated in our hospital for a period of time, they were transferred to lower - level medical institutions for further rehabilitation. These have led to a lack of data on hospital mortality and 90 - day mortality in our study. We plan to improve these deficiencies in subsequent research. We didn’t analyze specific treatment measures. Therefore, further research on the treatment timing and detailed plans for measures such as mechanical ventilation, glucocorticoids, CRRT, and plasma exchange is needed in the future. The regional onset of the disease and the rarity of the viral infection present challenges for external validation in larger or multicenter cohorts. Efforts will be made to collaborate with other research centers to strengthen the model’s reliability. The small sample size and single-center design may limit the generalizability and reproducibility of these results. Consequently, this study provides preliminary foundation to support future research.

In conclusion, severe SFTS patients face a high mortality risk, with GCS score, log10(virus loads), lactic acid, and CYS-C identified as IRFs. A nomogram based on these parameters shows considerable clinical value in predicting mortality, and thus can be used for early identification of patients with potential poor outcomes and improve clinical management. However, validation in a larger external cohort is essential for wider applicability.