Predicting the risk of mortality in children with dengue-induced hepatitis admitted to the paediatric intensive care unit

Abstract

BACKGROUND

Dengue-associated acute liver failure (PALF) accounts for a high mortality rate in children admitted to the pediatric intensive care unit (PICU). To date, there is a lack of data on clinical algorithms for estimating the risk of mortality in pediatric patients with dengue-induced severe hepatitis (DISH).

AIM

To determine the prevalence of PALF and identify the predictors of mortality among patients with DISH.

METHODS

This single-institution retrospective study was performed at a tertiary pediatric hospital in Vietnam between 2013 and 2022. The primary outcome was in-hospital mortality in pediatric patients with DISH, which was defined as either aspartate aminotransferase > 350 IU/L or alanine aminotransferase > 400 IU/L. Prognostic models for estimating the risk of death among patients with DISH were developed using a predefined set of clinical covariables and hepatic biomarkers on PICU admission and during the first 72 hours of admission. Area under the curve, multivariable logistic regression, and multiple imputation using the chained equation for missing values were performed. Backward stepwise model selection based on the Akaike information criterion was employed. Bootstrapping, calibration slope, and Brier score were used to assess the final models.

RESULTS

A total of 459 children with DISH were included in the analysis. The median patient age was 7.7 years (interquartile range: 4.3-10.1 years). The prevalence of dengue-associated PALF in children with DISH was 18.3%. Thirty-nine DISH patients developing PALF (8.5%) died. Hepatic biomarkers, including the international normalized ratio (INR) ≥ 2.11 and total serum bilirubin (≥ 1.7 mg/dL), showed high predictive values for mortality (all P values < 0.001). Multivariable models showed the significant clinical predictors of death from dengue-induced PALF in patients with DISH, including reduced level of consciousness (pain and unresponsive levels on the Alert, Verbal, Pain, Unresponsive scale), high vasoactive-inotropic score (> 30), and elevated levels of blood lactate, INR, and serum bilirubin. The final prognostic model demonstrated high discrimination, Brier score, and an acceptable calibration slope.

CONCLUSION

The prevalence of PALF in children with DISH is 18.3%. We developed robust prognostic models to estimate the risk of death in hospitalized children with severe dengue-induced hepatitis.

Key Words: Dengue; Dengue-associated acute liver failure; Dengue-induced severe hepatitis; Dengue shock syndrome; Hepatic encephalopathy; Mortality

Core Tip: The prevalence of dengue-associated acute liver failure in children with dengue-induced severe hepatitis (DISH) was 18.3%. The in-hospital mortality rate was approximately 8.5% among DISH patients developing acute liver failure. Hepatic biomarkers, including international normalized ratio (INR) (≥ 2.11) and total serum bilirubin (≥ 1.7 mg/dL), showed high predictive values for mortality. The significant predictors of mortality in children with DISH were decreased level of consciousness (pain and unresponsive levels on the Alert, Verbal, Pain, Unresponsive scale), high vasoactive-inotropic score (> 30), elevated blood lactate and INR levels during the first 24 hours of pediatric intensive care unit admission, and rising serum bilirubin during the first 72 hours of admission.

INTRODUCTION

Dengue-associated pediatric acute liver failure (PALF) is a critical complication in hospitalized children with severe dengue[1,2]. PALF primarily manifests as extensive hepatocellular necrosis and exacerbation of liver function, leading to hepatic encephalopathy and eventual death[3]. Our study cohort over the period 2013-2021 has shown that the prevalence of dengue-associated PALF in children with dengue shock syndrome was 4.2%[4]. Despite intensive interventions in the intensive care unit, the in-hospital fatality rate of dengue-associated PALF is high (68.3%)[5]. Our recent study revealed that combined therapeutic plasma exchange (TPE) and continuous renal replacement therapy (CRRT) significantly reduced the 28-day mortality rate of severe PALF to 39%[2]. Notably, early recognition, diagnosis, and appropriate intensive interventions at hospital admission may yield a better survival prognosis[5,6]. Our recent studies have also shown that better survival outcomes in pediatric intensive care unit (PICU)-admitted children with dengue-associated PALF are associated with rapid normalization of liver transaminase levels, metabolic biomarkers, coagulation profiles, and neurological status[1,2].

Several studies have reported that predictors of adult dengue PALF include low platelet counts, severe transaminitis, and elevated serum bilirubin levels[5,6]. Another Thai adult cohort study showed that End-Stage Liver Disease score ≥ 15 could predict the risk of PALF-associated death in adult patients with dengue-induced severe hepatitis (DISH); however, no specific predictor of PALF has been clearly determined[7]. To date, there are insufficient data on the prognosis of children with dengue-associated PALF[5,6]. Determining prognostic indicators for the risk of mortality in children with dengue-associated PALF is of utmost importance for categorizing and prioritizing treatments to improve survival and clinical outcomes. Therefore, we aimed to determine the prevalence of dengue-associated PALF among hospitalized patients with severe dengue-induced hepatitis and identify significant prognostic indicators for predicting the risk of death among patients with DISH.

MATERIALS AND METHODS

Study design, setting and participants

This single-center, retrospective study was performed at the Children’s Hospital 2, in Vietnam. All children with dengue who were admitted to the PICU between 2013 and 2022 were identified. Eligibility included age < 18 years, laboratory-confirmed dengue infection, and presence of DISH[7]. The exclusion criteria were a lack of serologically confirmed dengue infection, incomplete data on hepatic enzymes, including aspartate aminotransferase (AST) and alanine aminotransferase (ALT), and in-hospital mortality.

Primary outcome and candidate predictors

The primary outcome was death during hospitalization. A set of candidate predictors was preselected based on our clinical knowledge, disease pathogenesis, and established risk factors from previously published studies[5-8]. A set of predetermined covariates included age, sex, dengue severity, severe bleeding, neurological status [the Alert, Verbal, Pain, Unresponsive (AVPU) scale], high vasoactive inotropic score (VIS) (> 30), hematocrit level, platelet count, severe transaminitis, international normalized ratio (INR), serum sodium, total serum bilirubin, blood lactate, and creatinine.

Study definitions

Dengue infection confirmation was in accordance with the World Health Organization (WHO) criteria from 2009, based on positive test results for the non-structural 1 antigen or Dengue-IgM antibody test[9]. DISH was defined as baseline hepatic transaminase levels > 10-fold upper normal limit at PICU admission, and more specifically, AST > 350 IU/L or ALT > 400 IU/L)[7]. Severe bleeding was diagnosed according to the WHO dengue 2009 guidelines[9]. Severe transaminitis was defined as elevated AST and/or ALT levels ≥ 1000 IU/L[9].

Pediatric PALF was defined in accordance with the European Association for the Study of the Liver criteria, as an acute presentation of critical liver dysfunction with biochemical confirmation of hepatic injury in children without prior chronic hepatic diseases, coagulation disorders not improved by supplementing vitamin K, an INR level greater than 1.5, if patients have encephalopathy, or INR higher than 2.0 in the absence of encephalopathy[3].

The severity of dengue PALF was classified into three categories: (1) Mild; (2) Moderate; and (3) Severe. First, patients with mild PALF met the above criteria, with stable hemodynamics without vasoactive inotrope requirement, normal respiratory status, and a completely alert level on the AVPU scale. Second, patients with moderate PALF presented with dengue shock syndrome, mild respiratory failure with low-flow oxygenation support (nasal oxygen, and nasal continuous positive airway pressure), and alert and verbal response levels on the AVPU scale, without extracorporeal therapies. Lastly, patients with severe PALF had dengue shock syndrome and severe respiratory failure, requiring mechanical ventilation and extracorporeal therapies with either TPE or CRRT as indicated[1]. We commenced the combination of TPE and CRRT in patients with severe dengue-associated PALF, presenting with hepatoencephalopathy grades I and II showing poor treatment response or hepatoencephalopathy grades III and IV at any time in-hospital, high levels of ammonia (≥ 150-200 μmol/L), and serum lactate with a rising trend and poor therapeutic response[1,2].

Data collection

Clinical and laboratory data from the participants’ medical records were entered into well-structured case report forms (CRFs). In compliance with Good Clinical Practice, all patients’ names and related information were anonymized. Subsequently, data from the CRFs were entered into an electronic database for statistical analyses. The clinical outcomes of the patients were examined during the PICU stay and at discharge.

Data collection and measurements

Data on covariates were gathered during the first 24 hours of PICU admission, except for the highest level of total serum bilirubin within 72 hours of hospitalization. Blood and biochemical tests, including coagulation profiles, serum creatinine levels, and liver function tests, were performed using a machine (Alinity^® ci-series, Abbott, United States). The degree of hemodynamic support was evaluated by the VIS[10].

The in-hospital mortality was calculated from the time of admission to death. Dengue severity is redundant here, should be removed. The AVPU scale, with four simple categories, including alert, verbal response, response to pain, and unresponsiveness, was employed to assess the level of consciousness of the patients[11]. Notably, considering the correlation with the pediatric Glasgow Coma Scale (pGCS), the pain response has been shown to be equivalent to a pGCS from 4 to 12 points and unresponsive on the AVPU scale corresponding pGCS of 3-5 points[11].

Statistical analyses

Continuous variables are presented as median and interquartile (IQR) range and categorical data as numerical counts (n) and percentages (%). The primary study bias was missing information on patients’ hepatic biomarkers. The area under the curve (AUC) values, sensitivity, specificity, and accuracy were evaluated as indicators for predicting the risk of in-hospital death within 72 hours of PICU admission. Bivariate logistic regression with complete-case data and multivariable logistic analyses based on multiple imputation datasets were performed using the Hmisc package[12]. Predefined sets of covariates were chosen based on the disease pathogenesis and clinical experience. Notably, blood ammonia has been reported as an important predictor of death among patients with PALF; however, during the study period (2013-2022), serum ammonia tests were only performed in patients with moderate-to-severe PALF at our hospital. Therefore, serum ammonia was largely missing (90%) and we excluded this covariate from the analysis. Similarly, approximately 60% of the serum bilirubin data were missing and managed using multiple imputation methods[12]. Backward stepwise model selection and Akaike Information Criteria were used to identify the best predictive model, and interactions between the covariates were investigated. The internal validation model was performed using bootstrap resampling (n = 500). The performance of the prognostic models was evaluated using the C-statistic, calibration slope, and Brier score[13]. Statistical significance was set at P values < 0.05 for all statistical tests. The R statistical software (version 4.3.1, Boston, MA, United States) was used for all analyses.

RESULTS

Baseline characteristics of patients’ cohort on PICU admission

We identified 1971 children with severe dengue (dengue shock syndrome and dengue-induced liver disease) who were admitted to the PICU during the study period. A total of 459 patients experienced DISH and were included in the analysis (Figure 1). Of these, 84 patients developed DISH and fulfilled the criteria for PALF, while the remaining 375 patients did not develop PALF, progressing to clinical stability and complete recovery. Among 459 patients with DISH, 84 (18.3%) developed dengue-associated PALF. Of those diagnosed with dengue-associated PALF, 11 had mild, 24 had moderate, and 49 had severe life-threatening PALF that required TPE and CRRT (Figure 1).

Figure 1 The study flow-chart. PICU: Pediatric intensive care unit; DISH: Dengue-induced severe hepatitis; PALF: Dengue-associated acute liver failure; TPE: Therapeutic plasma exchange; CRRT: Continuous renal replacement therapy.

The clinical and laboratory data of the study participants upon admission to the PICU are presented in Table 1. The median age of the patients was 7.7 years (IQR: 4.3-10.1 years), and females accounted for 42% of the study participants. Thirty-eight patients (8.3%) with DISH had underlying diseases, including asthma, congenital heart disease, epilepsy, and β-thalassemia. Most patients experienced dengue shock syndrome (DSS) on the median day 5 (IQR: 4-5 day) after the onset of fever. Notably, compensated DSS was observed in 87.6% of the children, and the remaining 12.4% of the patients had decompensated DSS. Severe bleeding was observed in 76 (16.7%) patients. Systolic and diastolic shock indices were markedly elevated, and 45 patients (9.8%) had a high VIS (> 30). Complete blood counts revealed increasing hematocrit and low platelet counts. Significant hepatic transaminases, coagulation disorders, and elevated serum bilirubin levels were observed in participants experiencing DISH. The median blood lactate was 2.9 mmol/L (IQR: 2.0-4.5 mmol/L) on PICU admission.

Table 1 Baseline clinical and laboratory characteristics of study participants on pediatric intensive care unit admission and associated outcomes at discharge (n = 459), n (%).

Characteristics	No. participants	Statistics
Age (years)	457	7.7 (4.3-10.1)
Female patients	459	193 (42)
Underlying diseases	459	38 (8.3)
Day of occurrence of dengue shock since onset of fever (days)	451	5 (4-5)
Grading of DSS	459
Compensated DSS		402 (87.6)
Decompensated DSS		57 (12.4)
Severe bleeding	456	76 (16.7)
Respiratory rate (breaths/minute)	405	26 (24-30)
Systolic shock index (bpm/mmHg)	390	1.33 (1.15-1.6)
Diastolic shock index (bpm/mmHg)	390	1.83 (1.5-2.25)
Vasoactive-inotropic score > 30	459	45 (9.8)
White blood cell count, (× 109/L)	456	5.35 (3.76-7.8)
Peak hematocrit	446	47 (44-51)
Nadir hematocrit	446	38 (33-41)
Platelet cell count, (× 109/L)	457	32.6 (20-48)
Aspartate aminotransferase (IU/L)	459	751 (465-1416)
Alanine aminotransferase (IU/L)	459	344 (221-638)
International normalized ratio	406	1.40 (1.18-1.96)
Prothrombin time (second)	404	18 (15-27)
Activated partial thromboplastin time (second)	405	52 (43-69)
Peak serum bilirubin level (mg/dL)	119	1.9 (1.04-5.3)
Peak serum ammonia level (µmol/L)	64	117 (80-197)
Blood creatinine (µmol/L)	447	54 (44-63)
Troponin I (ng/mL)	164	0.036 (0.01-0.22)
Blood lactate (mmol/L)	293	2.9 (2.0-4.5)
Length of pediatric intensive care unit stay (days)	459	3 (2-5)
Length of hospital stay (days)	459	5 (4-10)
Fatal outcome	459	42 (9.2)

Summary statistics are presented as median (interquartile range) for continuous variables and frequency (%) for categorical variables. DSS: Dengue shock syndrome.

In-hospital mortality rate of patients with dengue-associated PALF

As shown in Figure 1, thirty-nine PICU-admitted children with DISH who developed PALF died during hospitalization, corresponding to a mortality rate of approximately 8.5%. Patients with severe PALF who experienced dengue shock syndrome, critical respiratory failure with mechanical ventilation support, and extracorporeal therapies with plasma exchange and/or CRRT accounted for a very high fatality rate, as noted in of the 32/49 patients (65%). In addition, other three DISH patients without PALF died of dengue-associated encephalitis, pulmonary hemorrhage, and multiorgan failure. Among the fatal cases, PICU-associated sepsis was notable, as 10/42 (23.8%) patients had a positive hemoculture with identified Acinetobacter baumannii and Pseudomonas aeruginosa.

Predictive values of associated serum biomarkers and in-hospital fatality

As presented in Table 2, two significant biomarkers, including INR during the first 24 hours of PICU admission and highest levels of total serum bilirubin levels within 72 hours of admission, showed high predictive values for in-hospital mortality (all presented significant P-values < 0.001). The optimal cut-off points for INR (≥ 2.11) and total bilirubinemia (≥ 1.7 mg/dL) revealed good predictive values for sensitivity, specificity, and accuracy.

Table 2 Area under the receiver operating characteristic curve for biomarkers and cut-offs for in-hospital mortality of children experiencing severe dengue-associated hepatitis.

Parameters	Area under the curve of mortality (95%CI)	P value	Optimal cutoff points of the parameters	Sensitivity	Specificity		Accuracy	Youden index
INR peak during first 24 hours of PICU admission	0.907 (0.858-0.956)	< 0.001	INR peak ≥ 2.11 during first 24 hours of admission	0.902	0.855		0.859	0.757
Peak of total serum bilirubin during 72 hours of PICU admission (mg/dL)	0.790 (0.704-0.875)	< 0.001	Peak of total serum bilirubin during the first 72 hours of PICU admission ≥ 1.7 (mg/dL)	0.941	0.565	0.672		0.506

INR: International normalized ratio peak; PICU: Pediatric intensive care unit.

Clinical predictors for in-hospital mortality in PALF

As presented in Table 3, bivariate analyses revealed important risk factors for in-hospital mortality in PICU-admitted children with DISH, including dengue severity, severe bleeding, significantly reduced mental status (pain response and unresponsive levels on the AVPU scale), high VIS (> 30), low platelet counts (< 20 × 10⁹/L), serum creatinine, INR, and the highest levels of total bilirubin, serum lactate, and ammonia within 72 hours of PICU admission. However, multivariate logistic analysis showed that the significant prognostic indicators of mortality in PICU-admitted children with dengue-associated PALF were altered mental status, as indicated by the pain response and unresponsive levels on the AVPU scale, high levels of VIS > 30, and blood lactate. No significant interactions were found between the covariates.

Table 3 Clinical predictors of in-hospital mortality in pediatric intensive care unit-admitted children with severe dengue-associated hepatitis from univariable and multivariable logistic analyses.

Factors	Non-survivors (n = 42)	Survivors (n = 417)	Unadjusted OR (95%CI), Pa	Adjusted OR (95%CI), Pb
Age (years)	7 (4.2-10)	7.8 (4.4-10.2)	0.98 (0.9-1.06), P = 0.55	-
Gender
Female	16 (38)	177 (42)	1.2 (0.62-2.3), P = 0.58	-
Male	26 (62)	240 (58)
Grading of DSS
Compensated DSS	29 (69)	373 (89)	3.8 (1.84-7.85), P < 0.001	-
Decompensated DSS	13 (31)	44 (11)
Severe bleeding
No	11 (26)	372 (89)	23.1 (10.9-49.1), P < 0.001	-
Yes	31 (74)	45 (11)
Levels of consciousness (on the Alert, Verbal, Pain, Unresponsive scale)
Alert and verbal response	15 (36)	387 (93)	23.2 (11.1-48.2), P < 0.001	4.43 (1.12-17.5), P = 0.03
Pain and unresponse	27 (64)	30 (7)
High vasoactive inotropic score (> 30)
No	7 (17)	407 (98)	204 (73-568), P < 0.001	39.2 (9.2-167), P < 0.001
Yes	35 (83)	10 (2)
Peak hematocrit (%)	46 (40-51)	47 (44-51)	0.95 (0.9-0.99), P = 0.024	-
Low platelet counts, (< 20 x 109/L)
No	25 (60)	319 (76)	2.21 (1.15-4.27), P = 0.018	-
Yes	17 (40)	98 (24)
Severe transaminitis
No	9 (21)	260 (62)	6.07 (2.83-13.0), P < 0.001	-
Yes	33 (79)	157 (38)
Serum sodium, (mmol/L)	128 (124-133)	128 (125-131)	1.0 (0.95-1.07), P = 0.74	-
Highest level of blood lactate (mmol/L)	10.3 (7.2-13.4)	2.4 (1.8-4.2)	1.67 (1.47-1.89), P < 0.001	1.61 (1.34-1.94), P < 0.001
Serum creatinine (µmol/L)	73 (58-123)	53 (44-61)	1.03 (1.02-1.04), P < 0.001	-

Summary statistics are median (interquartile) for continuous data and frequency (%) for categorical data. P^a and P^b withdrawn from complete-case analysis univariable and multivariable logistic analyses, respectively. DSS: Dengue shock syndrome; OR: Odds ratio.

Prognostic models for risk of death in severe dengue-induced hepatitis patients, with the combination of clinical predictors and hepatic biomarkers

Table 4 compares the prognostic values of the three predictive models for the risk of death in PICU-admitted patients with DISH. When clinically identified predictors were combined with hepatic biomarkers, the reduced models with either INR (prognostic model 1) or total serum bilirubin (prognostic model 2) significantly predicted the risk of death. However, the total bilirubin level was not statistically significant in the full model with clinically identified predictors, INR, and serum bilirubin level (prognostic model 3). These models were further evaluated using internal validation.

Table 4 Multivariable logistic models for predicting 28-day in-hospital mortality among dengue shock syndrome children with clinical predictors and biomarkers.

Prognostic factors	Model 1		Model 2		Model 3
	OR (95%CI), Pa		OR (95%CI),Pb		OR (95%CI), Pc
Pain and unresponsive levels on the Alert, Verbal, Pain, Unresponsive scale	13.86 (3.71-51.8)	< 0.001	14.36 (4.12-50.1)	< 0.001	13.96 (3.73-52.2)	< 0.001
High vasoactive inotropic score (> 30)	18.1 (4.99-65.4)	< 0.001	17.4 (5.06-60.2)	< 0.001	18.0 (4.98-65.3)	< 0.001
Highest level of blood lactate (mmol/L)	1.5 (1.27-1.78)	< 0.001	1.62 (1.37-1.91)	< 0.001	1.5 (1.27-1.78)	< 0.001
International normalized ratio levels, ≥ 2.11	11.3 (2.62-48.3)	0.001	_	_	9.1 (1.72-47.9)	< 0.01
Peak serum bilirubin level, ≥ 1.7 mg/dL	_	_	4.72 (1.12-19.8)	0.03	1.5 (0.29-7.81)	0.62

P^a, P^b, and P^c withdrawn from multivariable logistic analyses with multiple imputation data. Model 1: Clinical predictors and international normalized ratio (INR); Model 2: Clinical predictors and total serum bilirubin; Model 3: Clinical predictors, INR, and total serum bilirubin. OR: Odds ratio.

Performance of prognostic models for in-hospital death in patients with DISH and internal validation

The performance parameters of the three prognostic models were compared (Table 5). All reduced and full models revealed competency in performance as indicated by good discrimination with a high C-statistic index. Prognostic model 2 showed a slightly higher calibration slope than comparative models 1 and 3. However, the full prognostic model 3 had the least variability, as indicated by the lowest Brier scores in the training and test sets. Most importantly, as presented in Figure 2, full prognostic model 3 visually showed the best calibration plot, indicating the least variability. Based on the clinical importance of serum bilirubin in disease pathogenesis and the medical literature, it was retained in the final full prognostic model.

Figure 2 Calibration plots for the prognostic models of in-hospital mortality in pediatric intensive care unit-admitted children with dengue-induced severe hepatitis. A: Clinically predictors and international normalized ratio (INR); B: Clinically predictors and total serum bilirubin; C: Clinical predictors, INR and total serum bilirubin. The developed prognostic model showed relative consistency between the predicted values (x-axis) and the observed data (y-axis).

Table 5 Performance of prognostic models with internal validation.

Prognostic models	C-statistic	Calibration slope	Brier score
Model 1	0.974 (0.967)	1 (0.892)	0.021 (0.023)
Model 2	0.966 (0.960)	1 (0.906)	0.023 (0.027)
Model 3	0.975 (0.968)	1 (0.884)	0.020 (0.023)

Statistics are summarized for training (test) sets performed with bootstrap and internal validation. Model 1: Clinically developed model and international normalized ratio (INR); Model 2: Clinically developed model and total serum bilirubin; Model 3: Clinically developed model, INR, and total serum bilirubin.

DISCUSSION

The fatality rate of hospitalized patients with dengue shock syndrome ranges from 20% to 70%[5-8]. A dramatically elevated fatality rate has been reported in patients presenting with PALF secondary to prolonged shock, severe respiratory failure requiring mechanical ventilation, and extracorporeal therapies (CRRT and/or TPE)[1,2]. Apart from prognostic studies in adult cohorts, there are still inadequate data on the prognosis of children with dengue-associated PALF[5-8]. Therefore, in this study, we aimed to determine the prevalence of dengue-associated PALF in hospitalized children with DISH. We identified the most significant prognostic indicators for estimating the risk of mortality in patients with critical liver transaminases.

The reported incidence of dengue-associated PALF among the mild-to-severe dengue infected population in a large pediatric cohort in Thailand was 1.1%[5]. In this study, we showed that the prevalence of dengue-associated PALF in the pediatric population with DISH was as high as 18.3%. Notably, 39 of the 459 PICU-admitted children with DISH developed PALF and died during hospitalization, corresponding to a mortality rate of approximately 8.5%. These data are useful for establishing preemptive diagnostic schemes for dengue-induced PALF among hospitalized children presenting with severe transaminitis. Notably, there were 39 fatalities among the 84 patients diagnosed with PALF, indicating that the case fatality rate of dengue-induced PALF was 46.4%. This figure was markedly lower than that in previous reports from Thai pediatric (68.3%) and adult (58.8%) cohorts[5,7]. A possible explanation for the difference in mortality between our study and the Thai cohorts is that, in our study patients diagnosed with severe dengue-induced PALF were managed with extracorporeal therapies (TPE and/or CRRT). Most importantly, our research team has recently shown that combined TPE and CRRT interventions yield better survival outcomes in children with dengue-associated PALF and DSS[2]. In addition, our study had a larger sample size of dengue-associated PALF, indicating more statistical power than the Thai cohorts [5,7]. Currently, there is no classification for the severity of dengue-associated PALF, and in-hospital fatality rates vary greatly according to PALF severity. In this regard, we constructed a simplified categorization of PALF severity in which clinicians can rapidly classify severe patients and provide appropriate interventions.

To the best of our knowledge, this is to date the largest cohort of patients with dengue-associated PALF in which prognostic models to predict the risk of death among patients with DISH were developed. Akdogan et al[14] stated that patients with hepatic encephalopathy grades III and IV could have death rates of 42% and 100%, respectively. Notably, we have shown that improvements in hepatic encephalopathy, liver function tests, coagulation profiles, and biochemistry are associated with better survival outcomes in patients with dengue-associated PALF[1,2]. Therefore, we based this rationale on preselected all covariates in developing the prognostic models and identified the significant predictors of dengue-induced PALF. First, altered mental status can signify imminent hepatic encephalopathy among patients with dengue-associated PALF, which has been shown to be an independent predictor of death[15,16]. However, there are currently insufficient data on the correlation between decreased mental status and increased risk of death among patients with dengue-induced PALF. In this regard, we found that markedly reduced consciousness, as indicated by the pain and unresponsive levels on the AVPU scale, was a strong predictor of death in patients with dengue-associated PALF. Second, we have reported that a high VIS (> 30) and elevated blood lactate (≥ 4.2 mmol/L) are significant predictors of death in a pediatric cohort of 492 PICU-admitted patients with severe dengue[8]. In this study, these covariates were significant independent predictors of mortality due to dengue-induced PALF. This can be interpreted as the pathogenesis of acute liver failure involving massive hepatocyte necrosis, leading to multiorgan damage and hepatic encephalopathy, which clinically manifests as changes in consciousness and hemodynamics[17]. In addition, PALF can cause peripheral vascular dilation, microcirculatory vasoconstriction, and poor blood perfusion to organs, leading to injury and dysfunction in multiple organs[18]. Thus, a high VIS score and elevated serum lactate levels upon PICU admission, particularly an increasing trend in these parameters after admission, highly increase the risk of cerebral edema and death[1,2,8]. Nevertheless, previously published studies from Thailand have overlooked these critical factors when constructing predictive models for mortality from dengue-induced PALF[5-7]. In this respect, our study significantly fills the existing knowledge gap.

The primary mechanisms of hepatocyte injury and DISH involve direct cytopathic effects, immune responses, and reduced hemoperfusion in the liver[19]. Most importantly, rapid normalization of hepatic function, including liver metabolic biomarkers and coagulation dysfunction, is associated with better survival outcomes[1,2]. Therefore, we examined liver-related biomarkers for prognostication of death in patients with DISH. We found no statistical association between severe transaminitis and mortality risk, which is consistent with previous studies[7,8]. In terms of coagulation dysfunction, Teerasarntipan et al[7] revealed that an INR ≥ 1.5 had a good predictive value for estimating the risk of death from DISH, with an AUC of 0.83[7]. A higher INR > 3.5 was reported as a significant mortality predictor in acute liver failure according to King’s College criteria[20]. It was shown that INR levels ≥ 2.55 on hospital admission could predict the risks of death and/or liver transplant[21]. In addition, Akdogan et al[14] stated that INR values ranging from 1.5 to 5 were attributable to a death rate of 36%, and INR levels > 5 corresponded to a markedly increased fatality rate of 82%. In our study, the INR cut-off point ≥ 2.11 showed high prognostic values [sensitivity, 90.2%; specificity, 85.5%; odds ratio = 9.1 (95%CI: 1.72-47.9; P < 0.01)] among children with DISH. Our recent studies describing the dynamic variation in INR values in children with dengue-associated PALF revealed that INR rapidly increases with the occurrence of DISH[1,2]. Therefore, markedly increased INR upon admission is an important prognostic factor for in-hospital death in patients with dengue-induced PALF.

In dengue-induced PALF, the mechanisms of devastating hepatocytes can cause elevated serum bilirubin levels, including the direct cytopathic effects of the dengue virus, immune-mediated damage, decreased perfusion, hepatocyte necrosis, Kupffer cell hyperplasia and destruction, and inflammatory cell infiltration[21]. A total serum bilirubin level > 85.5 µmol/L (or 5 mg/dL) on admission can predict mortality[21]. Thus, we studied the total serum bilirubin levels in the predictive models of death in patients with DISH when combined with clinical predictors and INR parameters (Table 4). In the prognostic model 2, peak serum bilirubin level was a significant predictor of death from dengue-induced PALF; however, it became statistically insignificant in the prognostic model 3. Several possible explanations have been proposed for this finding. First, serum bilirubin levels have been reported to be highly variable in previous dengue cohorts[22-25]. This suggests that distinct mechanisms of liver injury in various cohorts could result in different changes in the serum bilirubin levels[21,26]. Second, because PALF can rapidly progress to critical multi-organ failure, prognostic indicators, including serum lactate and INR levels, can potentially show marked changes early in the occurrence of PALF[1,18,19]. In contrast, serum bilirubin elevation occurs late in the presence of massive hepatocyte necrosis and virus-induced cholangitis, causing vast blockage of the intrahepatic bile ducts[27]. Additionally, our recent study showed that serum bilirubin levels are commonly elevated since day 3 after PALF diagnosis[1]. From our real-life experience, patients with severe dengue-associated PALF may die soon after PICU admission, despite normal serum bilirubin levels on admission. It frequently takes several weeks to normalize serum bilirubin levels in dengue-associated PALF patients[1].

Notably, covariates reported as death indicators, including critical bleeding, low platelet counts, elevated blood creatinine levels, and dengue shock syndrome, were also examined[5,7,18,27]. However, these parameters were not statistically significant in the final multivariate model. Previous studies were restricted by relatively small sample sizes to determine the effect of these covariates in the predictive model of mortality from dengue-induced acute liver failure[5,7,18,27]. In addition, among the patients with DISH who died in-hospital, PICU-associated severe sepsis was found in 23.8% of these patients, and pathogens such as Acinetobacter baumannii and Pseudomonas aeruginosa were identified in the hemoculture. This finding indicates that secondary sepsis associated with PICU stay is also a significant contributing factor to the risk of death in the DISH population. These patients were likely susceptible to sepsis owing to invasive procedures such as catheterization and mechanical ventilation. According to our clinical supervision, these septic complications frequently occurred from days 4 to 5 after PICU admission. In this study, we aimed to develop a predictive model to estimate the risk of death among patients with DISH based on the covariates collected upon PICU admission. Nevertheless, secondary sepsis associated with PICU stay was also a notable risk factor for mortality in patients with DISH.

Our study had several limitations inherent to its single-center retrospective cohort design and unstandardized collection of clinical and laboratory data during hospital admission. The most significant bias was the missing information on serum ammonia, which was previously reported as an important prognostic indicator. However, this study retrospectively gathered data from medical records, and serum ammonia tests were only performed in patients with severe dengue-associated PALF. Similarly, the covariate total serum bilirubin was also markedly missing (approximately 60%), influencing the developed prognostic models at certain levels, despite being manipulated with multiple imputations. Considering these limitations, further investigations in prospective cohorts with appropriate study designs are essential to confirm our findings.

CONCLUSION

We developed robust prognostic models to estimate the risk of death in children with dengue experiencing severe hepatitis and identified significant predictors, including markedly reduced consciousness (pain and unresponsive levels on the AVPU scale), high VIS (> 30), rising blood lactate level, highly elevated INR, and serum bilirubin. The prognostic model will be externally validated in other cohorts.

ACKNOWLEDGEMENTS

We are grateful to the patients and administrative staffs, for their support with this study.