Dengue infection complicated by hemophagocytic lymphohistiocytosis: Experiences from 180 severe dengue patients.

Abstract Background Globally, ~500 000 people with severe dengue (SD) require hospitalization yearly; ~12 500 (2.5%) die. Secondary hemophagocytic lymphohistiocytosis (sHLH) is a potentially fatal hyperinflammatory condition for which HLH-directed therapy (as etoposide and dexamethasone) can be life-saving. Prompted by the high mortality in SD and the increasing awareness that patients with SD may develop sHLH, our objectives were to (1) determine the frequency of dengue-HLH in SD, (2) describe clinical features of dengue-HLH, (3) assess mortality rate in SD and dengue-HLH, and (4) identify mortality-associated risk factors in SD. Methods A 5-year retrospective single-center study in all adult patients with SD admitted to a tertiary intensive care unit in Malaysia. Results Thirty-nine of 180 (22%) patients with SD died. Twenty-one of 180 (12%) had HLH defined as an HLH probability ≥70% according to histo score (HScore); 9 (43%) died. Similarly, 12 of 31 (39%) fulfilling ≥4 and 7 of 9 (78%) fulfilling ≥5 HLH-2004 diagnostic criteria died. Peak values of aspartate aminotransferase (AST), alanine aminotransferase, lactate dehydrogenase, and creatinine correlated to fatality (odds ratios [ORs], 2.9, 3.4, 5.8, and 31.9; all P < .0001), as did peak ferritin (OR, 2.5; P = .0028), nadir platelets (OR, 1.9; P = .00068), hepatomegaly (OR, 2.9; P = .012), and increasing age (OR, 1.2; P = .0043). Multivariable logistic regression revealed peak AST (OR, 2.8; P = .0019), peak creatinine (OR, 7.3; P = .0065), and SOFA (Sequential Organ Failure Assessment) score (OR, 1.4; P = .0051) as independent risk factors of death. Conclusions Be observant of dengue-HLH due to its high mortality. A prospective study is suggested on prompt HLH-directed therapy in SD patients with hyperinflammation and evolving multiorgan failure at risk of developing dengue-HLH.

The global incidence of dengue infection has grown dramatically and has been estimated at 390 million infections yearly [1]. This mosquito-borne virus infection occasionally develops into a potentially lethal state called severe dengue (SD) [2]. Yearly, ~500 000 individuals with SD require hospitalization (~2.5% die), according to the World Health Organization (WHO) [2].
In Malaysia, 385 758 dengue infections were reported during 2000-2010 [23]. The Ministry of Health Malaysia reported 83 849 dengue infections in 2017, of whom 177 (0.21%) patients died. The Malaysian Registry of Intensive Care showed that dengue infection was among the top 5 diagnoses bringing patients to the intensive care unit (ICU) during 2013-2017. Prompted by the high mortality and that SD may develop into sHLH, we performed a 5-year retrospective single-center study in all adult patients with SD admitted to a tertiary care ICU with the objectives to (1) determine the frequency of dengue-HLH, (2) describe clinical features of dengue-HLH, (3) assess mortality rate, and (4) identify mortality-associated risk factors in SD. Interestingly, a recent meta-analysis on 122 dengue-HLH patients reported a fatality rate of 15% and highlighted the need to explore potential relationships between specific dengue findings and dengue-associated HLH [19].

METHODS
All adults (≥18 years) with SD and positive dengue NS1 antigen, and/or immunoglobulin (Ig) M (IgM) enzyme-linked immunosorbent assay (ELISA) and/or IgG ELISA [24] admitted to the multidisciplinary ICU at Hospital Sultanah Aminah, Johor Bahru, during 2010-2014 were included in the study, which was approved by the Medical Ethics and Research Committee, Malaysia. Severe dengue was defined according to the WHO 2009 classification, as follows: 1. severe plasma leakage leading to (a) shock or (b) fluid accumulation and respiratory distress; 2. severe bleeding (as evaluated by the treating physician); and/ or 3. severe organ involvement of (a) liver (aspartate aminotransferase [AST] and/or alanine aminotransferase [ALT] ≥1000 U/L), (b) CNS (impaired consciousness), and/ or (c) heart and other organs [20]. Impaired consciousness was in this study defined by confusion, seizures, drowsiness, aggressiveness, and/or altered behavior (excluding delirium and other non-dengue-related reasons).

Statistical Analysis
Statistical analysis was performed using IBM SPSS Statistics version 25 and R version 3.5.0. Descriptive statistics include frequencies, proportions, medians, and ranges. The distribution of continuous variables (AST, ALT, LDH, ferritin, triglycerides, fibrinogen, platelet count, creatinine, age on admission, APACHE II score, SAPS II score, and SOFA score) was assessed by histograms and normal qq-plots. Continuous variables were log-transformed using the natural logarithm, except for age and APACHE II, SAPS II, and SOFA scores. Fisher's exact test was used to assess association between mortality and different cutoff levels of AST and ferritin, respectively. Univariable logistic regression was used to evaluate risk factors for death. An odds ratio (OR) of 2.0 for a binary variable (ie, sex) indicates that the odds of dying are doubled and for a continuous variable that the odds are doubled for a 1-unit change in the variable (if log-transformed, a 1-unit change in its natural logarithm). The linearity assumption (log-transformed) of continuous predictors was checked by fitting models with restricted cubic splines using 3 knots (function rcs in R-package rms). A clear deviation from linearity was found for fibrinogen, which was excluded from multivariable regression due to many missing values (80%).
The parameters with the highest significance in univariable logistic regression were candidates for multivariable regression. For parameter pairs with high correlation (Pearson's correlation coefficient >0.6), only 1 parameter per pair was chosen to avoid multicollinearity. Candidates chosen for multivariable logistic regression were severe bleed, severe leak, severe organ involvement, age, SOFA score, AST, platelet count, and creatinine. Ferritin and HLH status were not included due to many missing values. The multivariable logistic regression model was fitted using backward regression, with Akaike's information criterion as selection criteria. Univariable logistic regression was used to evaluate having an HScore probability of HLH of 70% or greater and having 4 or more HLH-2004 criteria, respectively. Missing data were excluded listwise from the logistic regression models. In the multivariable logistic regression models, only complete cases were included. Sensitivity and specificity of the selected multivariable model were evaluated using a receiver operating characteristic (ROC) curve and assessing the area under the curve (AUC). An optimal cutoff was calculated as the value maximizing the sum of sensitivity and specificity.
Two-sample t tests were used when comparing logtransformed levels of AST, ALT, and ferritin between patients with and without corticosteroid treatment.
Of 8802 ICU admissions during the 5-year period, 287 had dengue infections but 9 had missing records, leaving 278 patients for analysis. Of these, 197 (71%) had SD. After excluding 17 patients aged less than 18 years, 180 patients were analyzed (103 men; 57%) ( Table 1). Most were NS1 positive (n = 137); of the rest, 37 were IgM and IgG positive, 1 IgG positive only, and 5 IgM positive only. A total of 108 patients had severe plasma leakage, 64 severe bleeding, and 99 severe organ involvement (Supplementary Table S1). Among the latter, 68 had severe liver involvement, 35 CNS involvement, and 38 involvement of heart and other organs (some had multiple organ involvement) (Supplementary Table S2). Altogether, 87 of 180 (48%) had severe involvement of the liver and/or CNS, a clinical picture resembling HLH.
Median nadir platelet count, lowest fibrinogen level, and median peak AST, ALT, LDH, and ferritin levels in HLH-2004≥4criteria patients and among HSprobability≥70 patients  are presented in Tables 2 and 3, as well as median APACHE II, SAPS II, and SOFA scores. Four patients with SD were also immunosuppressed, but only 1 patient, with lupus nephritis treated with azathioprine and prednisolone on admission, had an HScore probability of HLH of 70% or greater.

Parameters Related to Mortality
Clinical and laboratory findings in the 141 survivors and 39 nonsurvivors are detailed in Table 1. Peak values of AST, ALT, LDH, ferritin, and creatinine were all markedly higher in nonsurvivors, and platelet and fibrinogen nadir markedly lower. Furthermore, as expected, APACHE II, SAPS II, and SOFA scores within 24 hours of ICU admission were all markedly higher in nonsurvivors. Mortality in relation to various levels of ferritin and AST is presented in Supplementary Table S3. In univariable logistic regression, severe organ involvement (OR, 23.6; P < .0001), severe bleed (OR, 3.5; P = .00081), and severe leak (OR, 2.3; P = .042) all correlated with increased risk of death (Table 5). Peak AST, ALT, and LDH (log U/L) all strongly correlated with fatality (ORs, 2.9, 3.4 and 5.8, respectively; all P < .0001). Peak ferritin (log μg/L) (OR, 2.5; P = .0028), peak creatinine (OR, 32; P < .0001), lowest platelet levels (log × 10 9 /L) (OR, 1.9; P = .00068), and hepatomegaly (OR, 2.9; P = .012) were also mortality-associated risk factors, as was increasing age at hospitalization (by 5-year increment) (OR, 1.2; P = .0043). Finally, APACHE II, SAPS II, and SOFA scores all related to fatality (ORs, 1.3, 1.1 and 1.8, respectively; P < .0001 for all) (Table 5). Notably, increasing age was also positively correlated with these scores (data not shown).  In the multivariable logistic regression model, peak AST (log U/L) (OR, 2.8; P = .0019), peak creatinine (OR, 7.3; P = .0065), and SOFA score (OR, 1.4; P = .0051) turned out to be independent risk factors for death ( Table 6). The model had an AUC of 0.98 (Figure 2). Using a cutoff for the log-odds of 0.33 resulted in a sensitivity of 92% and a specificity of 96%.

DISCUSSION
Dengue cases have increased 30-fold in the past 50 years [20]. In our cohort, mortality in SD was 22% (39 of 180) and, notably, was ~40% in dengue-HLH. Peak values of AST, ALT, LDH, ferritin, and creatinine were mortality-associated risk factors, as were lowest platelet counts, hepatomegaly, severe organ involvement, severe bleed, severe leak, increasing age at hospitalization, and APACHE II, SAPS II, and SOFA scores (Table 5). Due to its high mortality, it is important to be observant of dengue-HLH and to consider HLH-directed therapy in affected patients.
To our knowledge, this report is the largest and most comprehensive on dengue-HLH in adults so far [11][12][13][14][15][16][17][18], except for a recent meta-analysis [19]. In children, there are 2 large studies on dengue-HLH; 1 from India reporting 23 children with HLH (≥5 of 8 fulfilled HLH-2004 criteria, including bone marrow hemophagocytosis) of 212 children with dengue infection [14] and 1 from Puerto Rico on 22 children who fulfilled ≥5 of 8 HLH-2004 criteria [17]. These reports, and the meta-analysis, support that HLH should be considered in rapidly deteriorating patients with SD with persistent fever, cytopenia, and markedly elevated ferritin and liver enzyme levels. Notably, ferritin has been reported to be valuable in identifying critically ill patients with sHLH, and even specifically predicting SD [27][28][29].
We found that the high dengue-HLH mortality makes it reasonable to consider a short course of HLH-directed therapy in selected patients. In line, prednisolone use has been significantly associated with less derangement in leukocyte and AST levels (P < .001 and P = .01, respectively) [30]. Importantly, HLH-directed treatment with dexamethasone and etoposide showed substantially reduced mortality in another potentially fatal viral infection associated with HLH (ie, severe EBV-HLH) [8][9][10]. Notably, T cells are infected in both EBV-HLH and acutely infected patients with dengue, and in dengue infection, T cells support viral replication and secrete viable virus particles. HLH-directed treatment efficacy in EBV-HLH may partly be due to lymphocyte reduction by corticosteroids and etoposide, which provides a rationale for a similar treatment approach in dengue-HLH [31,32]. Etoposide has been instrumental in reducing mortality in primary and secondary HLH [5][6][7][8][9][10]. A French study from a medical ICU concluded that there is a risk of missing the time frame when HLHdirected treatment may be effective [33]. In another study in 162 patients with HLH, etoposide included in first-line treatment tended to result in a better outcome (P = .079) [34]. In the dengue-HLH study from Puerto Rico, 16 of 22 (73%) patients were administered corticosteroids, 13 (59%) IVIG, 8 (36%) etoposide, and 8 (36%) "chemotherapy" (not defined further) [17]. Since only 1 child died, survival in etoposidetreated dengue-HLH was 7 of 8 (88%) or 8 of 8 (100%). A PubMed search on 15 April 2019 on "dengue, etoposide" revealed 2 additional reports where etoposide may have been used in dengue-HLH, but without specifics on outcome [35,36]. The exact mechanism of etoposide in hyperinflammation treatment is not established, but etoposide has been shown to substantially alleviate all symptoms of murine HLH, and the pharmacodynamics involved potent selective deletion of  activated T cells and efficient suppression of inflammatory cytokine production [37].
Of note, patients treated with corticosteroids in our study had more severe disease (higher peak AST, ALT, and ferritin levels; P < .0001). Nevertheless, 8 of 10 treated with dexamethasone survived. While these numbers are small, they are at least indicative and together with literature data suggest that early use of steroids in the context of HLH is worth further studies in SD [17,30], possibly with the addition of etoposide in patients with more pronounced hyperinflammation or rapidly deteriorating status, in particular patients with severe liver involvement (47% mortality). In our study, 25 of 39 (64%) of those who died, died within 3 days of ICU admission. Furthermore, they also manifested significantly higher levels of AST, ALT, LDH, creatinine, and ferritin than did survivors, which was particularly noticeable in dengue-HLH. Increasing peak ferritin and AST levels were associated with increasing fatality. In patients with dengue and with persistent fever and cytopenia along with severe organ involvement, a drastic increase in liver enzymes and/or altered mental state, in particular if ferritin is greater than 10 000 μg/L, clinicians need to consider HLH. Moreover, in infection-associated HLH in adults the etoposide dose is suggested to be reduced as compared with HLH-94/HLH-2004, from 150 mg/m 2 intravenous (iv) per dose to 50-100 mg/m 2 iv administered once weekly and only for a short period [38].
Our study has several limitations. First, patients may have died of SD and dengue-HLH in the region without being diagnosed (ie, fatalities may have been underestimated). Second, corticosteroid efficacy cannot be evaluated because it was not used in a structured manner. Third, laboratory results for ferritin, AST, ALT, and LDH were sometimes so high that diluted samples would have been necessary to obtain correct values (ie, sometimes the peak values reported were falsely low). Similarly, infrequent sampling may have resulted in missing peak and nadir levels. Fourth, making a diagnosis of secondary HLH is difficult. Fulfilment of 5 or more of 8 HLH-2004 criteria serves as a practical tool for HLH diagnosis [21], but these criteria were developed for children and are not validated formally for adults. In this study, 2 of 8 HLH-2004 criteria (NK-cell activity; soluble CD25) were not evaluated in any patient. Moreover, 3 additional criteria were evaluated in fewer than half of patients (hemophagocytosis, ferritin, hypertriglyceridemia/hypofibrinogenemia). Due to these limitations, we report both fulfillment of 4 or more and 5 or more of 8 HLH-2004 criteria [21], in addition to evaluation by HScore (retrospectively developed in a selected cohort of 312 adults) [22]. The reported best cutoff value for HScore was 169 points (~55% probability of HLH), corresponding to an accurate classification of 90% of patients. We used a higher level for probability of HLH (≥70%), to reduce overestimation of HLH. The frequent lack of diagnostic HLH parameters is an important study limitation. However, with more diagnostic parameters studied, the proportion of dengue-HLH could be even higher.
We conclude that it is important to be aware of dengue-HLH in SD due to its high mortality. Furthermore, a prospective study is needed in patients with, or at high risk of developing, dengue-HLH to evaluate prompt HLH-directed therapy with corticosteroids such as dexamethasone, and in very severe cases the possible addition of etoposide, as a complement to standard supportive management.

Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes
Author contributions. F. K. K. and C. C. T. provided study concept, methodology, data, investigations and analyses; C. C. T. also provided resources and F.  Receiver operating characteristic curve for the multivariable logistic regression model, including the variables age at hospitalization, SOFA score, peak aspartate aminotransferase, and peak creatinine. Abbreviations: AUC, area under the curve; SOFA, Sequential Organ Failure Assessment.