In-Hospital Mortality in Mechanically Ventilated Children With Severe Dengue Fever: Explanatory Factors in a Single-Center Retrospective Cohort From Vietnam, 2013–2022

Abstract

OBJECTIVES:

Severe dengue fever complicated by critical respiratory failure requiring mechanical ventilation (MV) contributes to high mortality rates among PICU-admitted patients. This study aimed to identify key explanatory variables of fatality in mechanically ventilated children with severe dengue.

DESIGN:

Retrospective cohort, from 2013 to 2022.

SETTING:

PICU of the tertiary Children’s Hospital No. 2, in Vietnam.

PATIENTS:

Two hundred children with severe dengue fever who received MV.

INTERVENTIONS:

None.

MEASUREMENTS AND MAIN RESULTS:

We analyzed clinical and laboratory data during the PICU stay. The primary outcome was 28-day in-hospital mortality. Backward stepwise multivariable logistic regression was performed to identify the explanatory variables associated with dengue-associated mortality at the initiation of MV. The model performance was assessed using C-statistics, calibration plot, and Brier score. The patients had a median age of 7 years (interquartile range, 4–9). Overall, 47 (24%) of 200 patients died in the hospital. Five factors were associated with greater odds of mortality: severe transaminitis (aspartate aminotransferase or alanine aminotransferase ≥ 1000 IU/L), high blood lactate levels, vasoactive-inotropic score (> 30), dengue encephalitis, and peak inspiratory pressure on MV. The model performance in training (test) sets was a C-statistic of 0.86 (0.84), a good calibration slope 1.0 (0.89), and a Brier score of 0.08.

CONCLUSIONS:

In our center, from 2013 to 2022, MV-experienced patients with severe dengue had a high mortality rate. The main explanatory factors associated with greater odds of death (related to critical liver injury, shock, and MV) may inform future practice in such critically ill patients.

RESEARCH IN CONTEXT.

• Patients with severe dengue requiring mechanical ventilation (MV) have a case fatality > 20%.

• Clinical data for this critical population are limited, but we have a curated dataset of such cases from our single-center in Vietnam, from 2013 to 2022.

• An improved understanding of the explanatory factors associated with death in MV supported dengue children may inform our future practice.

AT THE BEDSIDE.

• In our 2013–2022, single-center practice in Vietnam, the in-hospital fatality rate in patients with dengue requiring MV was 24%.

• At the start of MV support, we identified five explanatory variables associated with greater odds of mortality: severe transaminitis, high blood lactate levels, vasoactive-inotropic score (> 30), dengue-associated encephalitis, and increased peak inspiratory pressure.

• These data—liver involvement, shock state, and MV severity—may inform our future practice in severe dengue patients requiring MV.

In 2023, the World Health Organization (WHO) reported that there were 6.5 million patients with dengue and approximately 7300 associated deaths (1). Among the various severe manifestations of dengue, the known contributors to mortality include dengue shock syndrome (DSS), critical bleeding, dengue-associated encephalitis, acute liver failure, and severe respiratory failure requiring mechanical ventilation (MV) (2–4). Recent single-center, in-hospital and PICU retrospective studies, including our own work, indicate high mortality (up to one-in-four) in some settings (4–8). In this context, the current therapeutic approach is to make an early diagnosis and manage dengue complications in a timely manner, which includes prioritizing those at the highest risk. To this end, although some studies in adult patients with dengue have proposed prognostic models that assess mortality risk (9–11), a few data exist regarding the utility of such markers of severity in pediatric cohorts (6, 12, 13).

The recent approach in our 2013–2022 practice in Vietnam was to focus on the problem of acute respiratory failure in critically ill children with dengue (5, 8, 14, 15). The children requiring MV represent a special subset of dengue patients, and may account for much of the high in-hospital mortality in children with severe dengue (5, 8). In fact, this outcome may reflect progressive pulmonary interstitial leakage of plasma as part of the shock state. Therefore, we aimed to carry out an in-depth retrospective analysis of explanatory factors of mortality in our cohort of MV-experienced children with severe dengue.

MATERIALS AND METHODS

This study used a combination of datasets from our previous publications, including study cohort 1 (n = 112 MV patients) (14) and study cohort 2 (n = 119 MV patients) (15). Notably, the degree of overlap between the two studies was observed in 31 patients; therefore, the final dataset included 200 dengue-infected patients with MV in the study analysis. The aforementioned studies (14, 15) belonged to the primary project, namely the Vietnam Dengue-Infected Study, which aimed to build a large dataset of pediatric patients with DSS at tertiary Children’s Hospital No. 2, during 2013–2022. This project was ratified by the Scientific Committee and institutional review board (IRB) of the Children’s Hospital No. 2, Vietnam (IRB. No. 893/QD-BVND2 signed on June 6, 2022). Using secondary datasets from the primary studies placed less than minimal risk on participants. Hence, the requisite for informed consent was waived. This study strictly adhered to the principles of Good Clinical Practice and the ethical guidelines of the Declaration of Helsinki.

The eligibility criteria were younger than 18 years of age, laboratory-confirmed dengue infection, and acute respiratory failure requiring MV. The 2009 WHO criteria (16) were used to define the following: dengue infection by the nonstructural 1 (NS1) antigen test or the dengue-immunoglobulin M (IgM) antibody test, DSS, and severe organ impairments (i.e., respiratory failure, critical bleeding, kidney, and heart). Severe transaminitis was defined as aspartate aminotransferase or alanine aminotransferase levels 1000 IU/L or more. Dengue encephalitis was defined as presentation with altered mental state, including decreased consciousness or confusion, coma or altered behavioral consciousness more than 24 hours, and evidence of presence of dengue in the cerebrospinal fluid, that is, RNA, NS1 antigen or IgM, and monocytosis not explained by other cause (16–18). Severe bleeding was defined as significant blood loss causing hemodynamic instability, requiring urgent hemostatic intervention and blood product transfusion (16). Severe bleeding encompassed hemoptysis, hematemesis, melena, hematochezia, and hematuria.

Indications for MV in Children With Severe Dengue

Over much of the period of this cohort, 2013–2019, there was no international guidance detailing the indications and management of MV in children with severe dengue. However, at our center, our practice was based on the 2011 national dengue guidelines from the Vietnamese Ministry of Health Dengue Guidelines, as previously described in our 2024 report (8). Since 2019, we have based our practice on the 2019 national dengue guidelines (19). In summary, our initial approach for patients in respiratory difficulty is to use nasal continuous positive airway pressure. If this intervention fails, invasive MV is then used. Other indications for MV include critical dengue patients presenting with substantial pleural and abdominal cavity effusions, abdominal compartment syndrome, pulmonary edema and/or fluid overload, pediatric acute respiratory distress syndrome, and need for high-volume continuous IV infusion (≥ 7 mL/kg/hr) for 12 hours of PICU admission. Finally, MV is used to support patients with dengue encephalitis who are unstable on oxygen therapy alone or who present with apnea.

Measurements

The vasoactive inotropic score (VIS) was used to assess the degree of hemodynamic support for patients (20). The Pediatric Logistic Organ Dysfunction version 2 (PELOD-2) score was used to assess organ dysfunction (21).

Study Outcomes and Candidate Explanatory Variables

The primary outcome in this retrospective study was 28-day in-hospital dengue-associated mortality. This endpoint was calculated from the timing of hospital admission to the event of patient death during hospital stay. Predefined covariates in the outcome model gathered during the first 24 hours of PICU admission included the following: age, severity of DSS (i.e., decompensated and compensated), dengue-associated encephalitis, severe bleeding, severe transaminitis, peak hematocrit level, serum creatinine level, the highest level of blood lactate concentration, VIS at the start of MV, and the cumulative amount of fluid infused from referral hospitals and the first 24 hours during PICU admission. These covariates were predetermined based on our clinical experience and the medical literature (2, 3, 6, 12, 13).

Statistical Analyses

Summary statistics for the study variables are described using medians and interquartile ranges (IQRs) for continuous variables and numbers, percentages (%) for categorical data (Table S1, http://links.lww.com/PCC/C611). Complete case analyses were carried out because missing values were low: hematocrit (five cases), patient age (four cases), and peak inspiratory pressure (PIP) (three cases) (Fig. S1, http://links.lww.com/PCC/C611). Exploratory comparison of proportions is presented as mean difference and 95% CI. Regarding the predefined set of covariates, multivariable logistic regression was performed to identify the explanatory variables associated with 28-day, in-hospital, dengue-associated deaths. Backward stepwise selection used the Akaike Information Criterion (AIC), and covariates with p values of less than 0.2 were retained in the selection process. The best model was the one with the lowest AIC value. The least absolute shrinkage and selection operator (LASSO) was used to determine the best explanatory variables associated with the risk of dengue-associated fatality. Internal validation of the model was conducted using bootstrapping (n = 500). The performance of the model was assessed using C-statistics, area under the receiver operating characteristic curve, calibration-in-large, calibration slope, and Brier score. The significance level for all statistical tests was p values less than 0.05. All analyses were performed using the R statistical software (version 4.3.2; Boston, MA).

RESULTS

Over the period 2013–2022, we identified 200 children who were supported with MV for severe dengue at our hospital (Fig. 1). Thirty-three patients were intubated at the referring hospital before being transferred to our PICU. The median age was 7 (IQR, 4–9) years, and there were 105 (53%) girls (Table 1). Severe bleeding occurred in 84 (42%) of 200 patients, severe transaminitis in 82 (41%) of 200, and dengue encephalitis in 14 (7%) of 200. The other baseline clinical and laboratory characteristics of the cohort are summarized in Table 1. Elevated systolic and diastolic shock indices were slightly higher in the nonsurvivor than survivor group; however, the differences were not statistically significant. Complete blood cell counts showed elevated hematocrit levels and low platelet counts at presentation. In addition, a total of 33 patients were intubated at referral hospitals before transfer, slightly higher in the fatal group (9 [19%] of 47) than in the surviving group (24 [16%] of 153).

Figure 1.

The study flowchart of participants.

TABLE 1.

Clinical Characteristics and Outcomes of Patients During the First 24 Hours of PICU Admission

Characteristics	No. of Patients	All Patients (N = 200)	Nonsurvivors (n = 47)	Survivors (n = 153)
Age, yr	196	7 (4–9)	7 (4–10)	6 (4–9)
Female, n (%)	200	105 (53)	19/40	86 (56)
Body mass index, kg/m2	190	18.3 (15.9–21.8)	19 (17.4–22.3)	17.8 (15.4–21.6)
Dengue severity, n (%)	200			130 (85) 23 (15)
Compensated DSS		163 (82)	33/47
Decompensated DSS		37 (18)	14/47
Severe bleeding, n (%)	200	84 (42)	33/47	51 (33)
Severe transaminitis, n (%)	200	82 (41)	42/47	40 (26)
Dengue encephalitis, n (%)	200	14 (7)	5/47	9 (6)
Systolic shock index, beats/min/mm Hg	161	1.4 (1.2–1.7)	1.6 (1.3–1.9)	1.4 (1.2–1.7)
Diastolic shock index, beats/min/mm Hg	161	2.1 (1.7–2.7)	2.1 (1.9–2.9)	2.1 (1.7–2.7)
WBC count, × 109/L	198	6.6 (4.2–10.6)	7.1 (3.9–12.9)	6.5 (4.4–10.3)
Hemoglobin, g/dL	196	13.3 (11.4–15)	13.4 (10.6–15.1)	13.3 (11.6–14.9)
Peak haematocrit, %	196	46 (40–50)	46 (40–51)	46 (40–50)
Nadir haematocrit, %	196	36 (30–41)	33 (29–40)	36 (30–41)
Platelet cell count, × 109/L	200	31 (19–52)	25 (14–44)	32 (20–53)
International normalized ratio	192	1.8 (1.4–2.5)	2.5 (1.6–3.7)	1.8 (1.4–2.4)
Alanine aminotransferase, IU/L	200	229 (55–818)	848 (332–1,670)	126 (48–486)
Aspartate aminotransferase, IU/L	200	569 (160–2,153)	2,219 (708–4,458)	358 (126–1,205)
Serum creatinine, µmol/L	200	52 (42–68)	68 (56–118)	49 (40–62)
Blood lactate, mmol/L	173	3.1 (1.9–6.6)	7.2 (3.6–13)	2.5 (1.7–4.6)
Serum troponin I, ng/mL	119	0.08 (0.02–0.39)	0.33 (0.11–1.34)	0.05 (0.01–0.2)
Intra-abdominal pressure, cm H2O	123	29 (24–33)	30 (27–35)	28 (23–32)
Initial MV setting	200
Positive end-expiratory pressure, cm H2O		10 (8–12)	12 (10–15)	10 (8–12)
Peak inspiratory pressure, cm H2O		28 (22–34)	32 (24–36)	27 (22–33)
Vasopressor support, n (%)	200	109 (55)	33 (70)	76 (50)
Days of vasopressor support	200	2 (1–4)	5 (2–7)	1 (0–3)
Prior intubation from referral hospital, n (%)	200	33 (17)	9/47	24 (16)
Total fluid infused from referral hospitals and during the first 24 hr of PICU admission, mL/kg	200	97 (43–190)	140 (75–246)	82 (38–174)
PELOD-2 score  PELOD-2 on PICU admission  PELOD-2 at the initiation of MV  PELOD-2 at 24 hr after MV  PELOD-2 at discharge or death	197	6 (5–8) 7 (6–9) 7 (6–8) 2 (0–5)	10 (6–13) 11 (8–14) 11 (9–14) 21 (17–23)	6 (5–7) 6 (6–8) 6 (5–8) 1 (0–2)
Length of PICU stay, d	200	7 (5–9)	5 (2–8)	7 (5–9)
Length of hospital stay, d	200	12 (9–17)	5 (2–8)	14 (10–18)
Fatal outcome, n (%)	200	47 (24)	_	_

DSS = dengue shock syndrome, MV = mechanical ventilation, PELOD-2 = Pediatric Logistic Organ Dysfunction 2 score.

Summary statistics are median (interquartile range) for continuous variables and frequency (%) for categorical variables.

Dashes indicate null values.

Regarding the baseline data by survival grouping, there are a number of potential explanatory factors associated with mortality rather than survival (Table 1). For example, mortality vs. survival was associated with a higher proportion of patients with severe transaminitis (42 of 47 vs. 40 of 153; mean difference 63% [95% CI, 48.8–71.8%], p < 0.001) and severe bleeding (33 of 47 vs. 51 of 153; mean difference 37% [95% CI, 20.8–50.2%], p < 0.001). Other associations with mortality in the exploratory data summary included blood biochemistry (see creatinine and lactate concentrations) and PELOD-2 scores (Fig. S2, http://links.lww.com/PCC/C611).

The median volume of fluid infused from referral hospitals and within 24 hours of PICU admission was 97 (IQR, 43–190) mL/kg. Over half (109 [55%] of 200) of the cohort required vasopressor support, and the average duration of vasopressor support was 2 (IQR, 1–4) days. Survival (compared with time to death) was associated with a longer PICU stay (7 [IQR, 5–9] vs. 5 [IQR, 2–8] d, p = 0.03) and shorter duration of vasopressor support (1 [IQR, 0–3] vs. 5 [IQR, 2–7] d, p < 0.01).

28-Day In-Hospital Mortality

There were 47 in-hospital deaths in this cohort (Table 1). Using the data from the start of MV support, Table 2 is a summary of the univariate and multivariable logistic analyses with backward stepwise multivariable logistic regression (AIC criteria and LASSO) (Figs. S3 and S4, http://links.lww.com/PCC/C611). The explanatory factors associated with greater odds of dengue-associated mortality in patients supported with MV during the first 48 hours of PICU admission were severe transaminitis, dengue-associated encephalitis, high blood lactate levels and VIS (> 30), and increased PIP.

TABLE 2.

Explanatory Factors at the Initiation of Mechanical Ventilation in Models for In-Hospital Mortality Caused by Severe Dengue

Candidate Factors	Full Modela		Final Modela
	OR (95% CI)	p	OR (95% CI)	p
Severe prolonged dengue shock syndrome	3.56 (0.75–16.9)	0.11	—	—
Severe bleeding	2.67 (0.82–8.7)	0.10	—	—
Severe transaminitis	6.71 (1.6–29)	0.01	8.95 (2.23–35.9)	< 0.01
Dengue encephalitis	71 (5.9–862)	< 0.001	19.5 (2.42–156.8)	< 0.01
Peak hematocrit level, %	1.07 (1.0–1.15)	0.06	—	—
Low platelet cell count (< 20 × 109/L)	1.15 (0.35–3.83)	0.82	—	—
Log-2 blood lactate, mmol/Lb	2.93 (1.54–5.6)	0.001	2.9 (1.7–4.93)	< 0.001
Serum creatinine, µmol/L	0.99 (0.99–1.01)	0.84	—	—
Prior intubation from referral hospital	0.69 (0.15–3.12)	0.62	—	—
Peak inspiratory pressure on mechanical ventilation (per + 1 cm H2O)	1.09 (1.01–1.16)	0.02	1.07 (1.01–1.14)	0.02
Log-2 Cumulative fluid infused from referral hospitals and within 24 hr-PICU admission, mL/kgb	0.89 (0.59–1.33)	0.56	—	—
Vasoactive inotropic score (> 30)	3.98 (1.19–13.3)	0.03	3.25 (1.1–9.62)	0.03

OR = odds ratio.

^aMultivariable logistic regressions based on complete-case analyses.

^bStandardization of covariates by log-2 transformation.

Dashes indicate insignificance in the final prognostic model.

The final mortality outcome model, with five explanatory factors (i.e., approximately nine outcome events per factor), was further evaluated for discrimination and calibration. Overall, there was good discrimination between survivors and nonsurvivors, with C-statistics of 0.86 in the training set, and 0.84 in the test set (Fig. 2). There was a good level of consistency between the predicted and observed mortality data, as shown with good calibration-in-the-large 0 (–0.01) and calibration slope in the training data of 1.0 (test data 0.89), and the Brier score was 0.08.

Figure 2.

The calibration plot showing good consistency between the model values and actual observational data, indicating high calibration slope and intercept, and a satisfactory Brier score.

DISCUSSION

In this retrospective cohort of MV in children with severe dengue managed at our center in Vietnam, from 2013 to 2022, we have made three key observations beyond those in our previous analyses (4, 8, 14, 15). First, the in-hospital mortality was 47 of 200 (24% [95% CI, 18–30%]) at our institution. Second, we identified five key explanatory variables associated with greater odds of mortality in patients with severe dengue at the time of starting MV. Third, we believe that these results may enable us to refine our intervention strategies in critical dengue in children.

The in-hospital mortality of 24% (95% CI, 18–30%) of MV children with dengue at our institution, 2013–2022, is consistent with mortality in a contemporary (2016–2019) pediatric dengue cohort in New Delhi, India (6). In our current report, after some preliminary exploratory analyses, we developed a multivariable model that highlighted five key explanatory variables associated with greater odds of mortality in cases of severe dengue at the time of initiating MV: transaminitis, dengue-associated encephalitis, high blood lactate levels and VIS (> 30), and increased PIP. These factors have been previously identified as factors associated with mortality in studies independent to our own (3, 6, 22). In terms of the use of MV support in our series, transaminitis reflects liver involvement, which we have previously described in the 2013–2021 cohort (4, 15). Furthermore, our observation that dengue-associated encephalitis is an explanatory variable associated with greater odds of mortality adds to the recent literature on the spectrum of fatal CNS complications of dengue (23).

Regarding the shock variables—high blood lactate concentration and VIS > 30, and high PIP, and the associated greater odds of mortality—there is likely a complex interaction in the underlying pathophysiology, which we have discussed in a prior publication about our 2013–2022 cohort (15). On the one hand, there is the role of fluid resuscitation and interstitial fluid leak, including the development of pulmonary edema necessitating more intensive respiratory support. Of note, two-thirds of our patients were transferred from provincial hospitals and had already received large volumes of resuscitation fluid. These referred patients accounted for a larger proportion of participants with huge plasma leakage, disseminated edema, and more critical respiratory failure. Even though we failed to identify an association between cumulative fluid volume administered during the first 24 hours of PICU admission and greater odds of mortality, we do recognize that it is difficult to accurately assess the fluid resuscitation variable in our study population. The problem is the insufficient data on the amount of fluid infused prereferral in a significant number of patients. On the other hand, separate from fluid accumulation and the lungs—as we have reported previously in our 2013–2021 cohort using point-of-care ultrasound (POCUS) (8)—there is a thoraco–abdominal–cardiac interaction to consider. For example, patients with severe DSS frequently present with disseminated body edema and respiratory failure, necessitating MV with elevated positive end-expiratory pressure (PEEP). Positive-pressure MV can aggravate the clinical status of patients with DSS and increase intra-abdominal pressure (IAP) (23–25). Also, high positive pressure in MV can cause dengue-associated obstructive shock syndrome (26). As a consequence, decreased preload, reduced cardiac function, and multiple organ failure, all with an increased lactate-to-bicarbonate ratio is associated with an over eight-fold greater odds of mortality in our published cohort, in which 119 of 492 patients with dengue underwent MV (15).

Taking all of the above together, the final issue is whether these data enable us to refine our intervention strategies in critical dengue. To date, our collected studies using the cohort of children with dengue managed at our institution from 2013 to 2022 (4, 8, 14, 15) translate into the following management scheme: 1) continued fluid resuscitation to ensure preload under POCUS guidance; 2) abdominal paracentesis for decompression of IAP; and 3) administration of vasopressors to sustain adequate preload. In this context, first, the initial goal of resuscitation in critical dengue is to use the minimum fluid volume necessary to maintain adequate tissue perfusion (16). When hydrostatic pressures are high in the abdominal and thoracic cavities, we have reported in our cohort that POCUS-guided fluid resuscitation is appropriate for severe prolonged DSS with significant preload deficit and complex hemodynamics (8). However, further independent studies are required to externally validate our protocol. Second, like others (11), we have used timely peritoneal decompression when abdominal compartment syndrome occurs, that is, when abdominal pressure exceeds 27 cm H₂O, and is accompanied by progressive organ damage or the emergence of new organ damage (8). Third, vasopressors improve venous return, restore preload, and increase cardiac output, in patients with prolonged DSS and multiple organ damage (13). In our cohort, we have reported using inotropes if there is decreased myocardial contractility with an ejection fraction less than 50% (8), and there is no surprise that a VIS > 30 is associated with greater odds of mortality, as we have previously found in our practice (15). However, perhaps as we recently reported, the lactate-to-bicarbonate ratio may be a variable to monitor (15).

Next, in relating the fluid strategy to the specifics of ventilatory management in DSS patients with intra-abdominal hypertension (IAH), there are a number of factors to consider (24–26) . For example, a 2020 Intensive Care and Emergency Medicine algorithm suggests that using a low tidal volume (Vt) strategy and optimized PEEP enhances lung compliance (27). Plateau pressure (P_plat) on MV should be adjusted relative to the patient’s IAP, to ensure adequate perfusion to the vital organs. In addition, PEEP adjustments, measured in cm H₂O, should approximate the IAP (in mm Hg) and not exceed 15 cm H₂O to prevent a reduction in cardiac output. Notably, an airway driving pressure (DP) > 14 cm H₂O is associated with poor outcomes in MV patients (28–30). Patients with IAH often present with high pleural and thoracoabdominal pressures, which require P_plat > 30 cm H₂O and DP > 14 cm H₂O to ensure adequate transpulmonary pressure to maintain the minimal Vt of 6 mL/kg (31). In our study, the median IAP was 29 (IQR, 24–33) cm H₂O, which surpassed the suggested PEEP threshold of 15 cm H₂O. Therefore, our current approach is to use PEEP only to levels that do not affect cardiac output or visceral perfusion; PIP is increased to maintain low Vt and, if needed, abdominal decompression is repeated.

Finally, our single-center, retrospective 2013–2022 cohort of pediatric patients with critical dengue requiring MV support has some limitations inherent to the study design. Overfitting may occur when developing the full model, regarding more covariates vs. limited outcome events. Nevertheless, the final model with the five most significant explanatory factors was stringently penalized using the LASSO method. Most importantly, the current report must be read in the context of the four other works published from the cohort (4, 8, 14, 15), as recommended by Pediatric Critical Care Medicine (32). That said, we have been able to identify patient characteristics associated with greater odds of mortality in children with critical dengue requiring MV, such as severe transaminitis, dengue-associated encephalitis, high blood lactate levels and VIS (> 30), and increased PIP. These results also provide insights that may inform our future practice when using MV in DSS.

ACKNOWLEDGMENT

We thank all the administrative staff, physicians, nurses, and patients for their participation and contribution in this study.