Clinical Characteristics and Predictors of Mortality in Critically Ill Influenza Adult Patients.

Severe influenza is associated with high morbidity and mortality. The aim of this study was to investigate the factors affecting the clinical outcomes of critically ill influenza patients. In this retrospective study, we enrolled critically ill adult patients with influenza at the Kaohsiung Chang Gung Memorial Hospital in Taiwan. We evaluated the demographic, clinical, and laboratory findings and examined whether any of these measurements correlated with mortality. We then created an event-based algorithm as a simple predictive tool using two variables with statistically significant associations with mortality. Between 2015 and 2018, 102 critically ill influenza patients (median age, 62 years) were assessed; among them, 41 (40.1%) patients died. Of the 94 patients who received oseltamivir therapy, 68 (72.3%) began taking oseltamivir 48 hours after the onset of illness. Of the 102 patients, the major influenza-associated complications were respiratory failure (97%), pneumonia (94.1%), acute kidney injury (65.7%), adult respiratory distress syndrome (ARDS) (51%), gastrointestinal bleeding (35.3%), and bacteremia (16.7%). In the multivariate regression model, high lactate levels, ARDS, acute kidney injury, and gastrointestinal bleeding were independent predictors of mortality in critically ill influenza patients. The optimal lactate level cutoff for predicting mortality was 3.7 mmol/L with an area under curve of 0.728. We constructed an event-associated algorithm that included lactate and ARDS. Fifteen (75%) of 20 patients with lactate levels 3.7 mmol/L and ARDS died, compared with only 1 (7.7%) of 13 patients with normal lactate levels and without ARDS. We identified clinical and laboratory predictors of mortality that could aid in the care of critically ill influenza patients. Identification of these prognostic markers could be improved to prioritize key examinations that might be useful in determining patient outcomes.


Introduction
Influenza is an acute viral respiratory infection caused by different types of influenza viruses: influenza A, B, and C [1]. Influenza A subtypes H1N1, H3N2, and influenza B are the most common causes of human influenza [1]. The illness is usually mild and characterized by a sudden onset of fever, cough, sore throat, runny nose, headache, myalgia, and malaise [1]. However, the virus can cause severe illness or even death, especially in high-risk individuals such as young children, the elderly, patients with certain comorbid chronic diseases, and immunocompromised patients [2,3]. Annually, the World Health Organization estimated that global influenza epidemics result in 3 to 5 million cases of severe illness and 290,000 to 650,000 deaths [4,5]. In 2009, a swine-origin influenza A (pandemic 2009 A/H1N1; pdm09 A/H1) emerged and rapidly caused a global pandemic [6]. Remarkably, from 12 April 2009 to 10 April 2010, there were 60.8 million cases and 12,469 deaths in the United States due to pdm09 A/H1 [7].
Severe complicated influenza has a significantly high mortality and morbidity [8,9]. Secondary bacterial pneumonia and acute respiratory distress syndrome (ARDS) are some of the common pulmonary complications of influenza, often followed by grave outcomes [10]. In addition to pulmonary complications, extra-pulmonary complications such as meningoencephalitis, myocarditis, and rhabdomyolysis have also been reported to be associated with either influenza A or B [11][12][13]. While early antiviral therapy may reduce complications of influenza [14][15][16], the majority of patients tend to delay seeking medical care and timely diagnosis, leading to the subsequent development of influenza-associated complications, particularly in the elderly and those with comorbid illnesses. Hence, key clinical data are crucial pieces of information that can help clinicians deliver the necessary management to critically ill influenza patients in a timely manner. In the present study, we reviewed the entire clinical course and laboratory data of critically ill patients with influenza and then explored the risk factors independently associated with death as well as constructed a valuable event associated algorithm to help in the assessment of risk of death.

Study Design and Patients
We retrospectively studied all critically ill adult patients (aged ≥18 years) with laboratoryconfirmed influenza infection admitted between 2015 and 2018, at the Kaohsiung Chang Gung Memorial Hospital, a 2700-bed primary care and tertiary referral medical center that included 5 adult medical and 4 surgical intensive care units (ICUs) in Taiwan. Included critically ill influenza patients were those admitted to an ICU, or those with complications of lower respiratory tract infection and/or multiorgan failure, or those requiring mechanical ventilation during hospitalization. Individuals were excluded from the study if they are <18 years of age; if they are treated as outpatients; or if they had mild flu-like symptoms. Confirmation of influenza virus infection required a positive finding in the respiratory specimen (nasopharyngeal swab and/or pharyngeal swab) by one or more of the following methods: rapid influenza diagnostic test (Formosa One Sure Flu A/B Rapid Test Kit), isolation of the virus in tissue-cell culture (Madin-Darby canine kidney (MDCK) cell line), or reverse-transcriptase-polymerase chain reaction (RT-PCR) (QiAamp Viral RNA Mini Kit; TAIGEN Bioscience Corporation, Taiwan). The choice of diagnostic test (rapid influenza diagnostic test, virus isolation or RT-PCR) for confirming influenza was based on the individual physicians' judgment.

Data Collection and Definitions
A standardized form for clinical data collection was designed. The data were mainly retrieved from the hospital's electronic medical records and were supplemented by a secondary manual search. The following data were collected: demographic characteristics, underlying medical conditions, clinical signs and symptoms, antiviral treatment course (oseltamivir or peramivir therapy), results of laboratory tests and radiography findings at the time of presentation and during the entire clinical course, in-hospital complications, and fatality.
Acute respiratory failure was defined as arterial partial pressure of oxygen (PaO 2 ) <60 mmHg in ambient air, or tachypnea >30/min. ARDS was defined as acute respiratory distress characterized by bilateral pulmonary consolidation and severe hypoxemia (PaO 2 /Fraction of inspired oxygen ratio <300 mmHg) in the absence of evidence for cardiogenic pulmonary edema [17]. Fulminant hepatitis meant alanine aminotransferase levels (ALT) greater than 16.6 µkat/L. Acute kidney injury was defined as a rapid increase in the serum creatinine level to >44.2 µmol/L compared with that at presentation. Rhabdomyolysis was defined as a five-fold increase in the serum concentrations of creatine phosphokinase above the upper limit of the normal range (reference value, 0.2-2.2 µkat/L), with >95% creatine phosphokinase-muscle fraction. Meningoencephalitis was defined as an altered mental status, fulfilling at least 2 of the following criteria: (1) fever, (2) seizure, (3) focal neurologic signs, (4) abnormality of cerebrospinal fluid, (5) neuroimaging suggestive of encephalitis, and (6) abnormal findings on electroencephalography consistent with encephalitis [18]. A galactomannan cutoff optical density index of >0.5 was used to define positivity for serum and bronchoalveolar lavage samples [19,20]. Mortality was defined as death occurring during the hospital stay for influenza.

Statistical Analysis
To analyze the predictors of mortality among critically ill influenza patients, we initially compared demographic, clinical characteristics, and laboratory findings as well as complications of survivors and nonsurvivors using Fisher's exact test for categorical variables and Mann-Whitney U test for numerical variables. The differences were considered significant at p < 0.05. Significant variables in the univariate analyses were entered into a multivariate logistic regression model to identify independent predictors of mortality in critically ill influenza patients. We used receiver operating characteristic curves (ROC) to select cutoff points for independent numerical predictors according to visual assessment of the highest sensitivity and specificity. We then created an event-based algorithm as a simple predictive tool using 2 variables with statistically significant associations with mortality (p value less than or equal to 0.001) in the multivariate model. Data were entered and analyzed using the Statistical Package for the Social Sciences statistical software (version 19.0; SPSS Inc., Chicago, IL, USA).

Event-Based Algorithm
The median highest lactate values (reference value <2.1 mmol/L) and median time from presentation to the highest lactate level among survivors and nonsurvivors were 2.7 mmol/L and 4.3 mmol/L and three days and four days, respectively. We selected the lactate variable in multivariate analyses and plotted the ROC curve to identify the optimal cutoff value for predicting mortality. The optimal cutoff of lactate level for predicting mortality was 3.7 mmol/L with an area under curve of 0.728, and the sensitivity and specificity of this cutoff were estimated at 63.9% and 74.1%, respectively (Figure 1). We then created an event-associated algorithm that included lactate level and ARDS-the two variables that were independently and significantly associated with death in multivariate analyses ( Figure 2). Fifteen (75.0%) of the 20 patients with lactate levels of 3.7 mmol/L or above and with ARDS died, compared with only one (7.7%) of 13 patients with lactate levels below 2.1 mmol/L and without ARDS (p < 0.001) and 1 (10%) of 10 patients with lactate levels between 2.1 mmol/L and 3.7 mmol/L and without ARDS (p < 0.001).

Discussion
In a study involving 444 adult patients with influenza in hospitals in the United States, the mortality rate was 20.9% [21]. Furthermore, a mortality rate of 20.6% was reported by Francisco et al. in their study of 2059 patients admitted to ICUs for influenza infection [22]. In our study, a mortality rate as high as 40% was found in 102 critically ill adult patients with influenza. However, which variables can predict poor patient outcomes after influenza virus infection remain to be elucidated. In the present study, our dataset included clinical signs and symptoms and laboratory results at presentation and the entire course of hospitalization as well as complications during the clinical course. We determined which demographic, clinical, and laboratory findings were associated with death that could help clinicians deliver timely and sufficient treatment to critically ill influenza patients. Our results underscore that high blood lactate levels during hospitalization, ARDS, acute kidney injury, and gastrointestinal bleeding were independent risk factors of mortality in critically ill influenza patients.
High blood lactate levels indicate tissue hypoxia due to increased lactate generation via anaerobic glycolysis [23]. High blood lactate levels have been correlated with poor outcomes in patients with bacterial sepsis and septic shock [24]. In the present study, high blood lactate levels were found to be significantly and independently associated with fatal outcomes in critically ill influenza patients. In addition, nonsurvivors had a significantly higher prevalence of acute kidney injury and gastrointestinal bleeding and received ECMO treatment in our series. Importantly, acute kidney injury and gastrointestinal bleeding have been shown to be independent risk factors of mortality. We believe that these complications are caused by clinicians' lack of awareness of early detection of organ hypoperfusion. As patients in early phases of hypoperfusion do not always show obvious clinical signs, blood lactate level may be an important marker for this disorder. Thus, timely recognition of organ hypoperfusion and initiation of effective volume replacement to reverse tissue hypoxia are critical steps in preventing mortality and morbidity. Notably, the median time interval from patient arrival to measurement of highest blood lactate was four days in nonsurvivors in our series. Further, the median time from illness onset to fatality was 18.5 days. This finding indicates that blood lactate levels can be a useful early marker assisting clinicians in predicting the outcomes in critically ill influenza patients.
Our series here showed gastrointestinal bleeding occurred in one-third of included patients and more than half of them with fatal outcome. Stress ulcer prophylaxis for critically ill patients is not universally practiced in our series. The causes of gastrointestinal bleeding in these patients are multifactorial. Our study underlines that clinicians should be alert to possible gastrointestinal bleeding when caring for a critically ill influenza patient because this complication potentially leads to death if it is not recognized early and treated accordingly.
ARDS is a lethal complication of influenza infection [25]. Ortiz et al. estimated that the incidence of influenza-associated acute respiratory failure was 2.7 events per 100,000 person-years [26]. In a study of 58 patients with ARDS, 28 (48.2%) were due to influenza virus infection, and 32.1% of the patients with influenza-associated ARDS received ECMO treatment [27]. Davies et al. reported that the incidence of pdm09 A/H1-associated ARDS sufficient to warrant consideration of ECMO was estimated at 2.6 cases per million population [28]. ARDS is an independent risk factor for hospital mortality in critically ill influenza patients, and the mortality rate can be as high as 52% [29]. The present study results are consistent with previous findings wherein 97% of critically ill influenza patients developed acute respiratory failure with a median time of four days between illness onset and respiratory failure, and 51% of them subsequently developed ARDS during their clinical course. In addition, approximately one-third of the patients with influenza-associated ARDS required ECMO for profound hypoxemic respiratory failure. Our study highlights that severe oxygenation failure occurred rapidly after hospital admission and that clinicians should not delay delivering appropriate rescue therapies as well as deploying ICU resources to meet this treatment requirement, particularly during the influenza epidemic.
In the present study, we established a simple event-associated algorithm including blood lactate level and ARDS for timely detection of critically ill influenza patients who are at greater risk of mortality. Notably, critically ill influenza patients without ARDS but with a blood lactate concentration of 3.7 mmol/L had an in-hospital mortality of 47.1%, and more importantly, the mortality rate increased to 75% for those with high blood lactate (≥3.7 mmol/L) and those that developed ARDS. In contrast, only 7.7% of critically ill influenza patients without ARDS and with normal blood lactate levels died. Considering the high mortality rate among critically ill influenza patients, this event-based algorithm could aid in the timely decision-making process and provision of prompt intensive care for patients with potentially fatal outcomes, particularly in resource-limited areas; some key laboratory tests such as blood lactate might be of greater value than others when allocating limited healthcare resources.
Previous studies have shown that early administration of an antiviral agent is associated with a shorter duration and reduced severity of illness [14][15][16]. Greater benefits were shown with early treatment initiated within two days after the onset of illness [30][31][32]. In our study, the median time from illness onset to hospital presentation was three days and more than two-thirds of the patients received delayed (48 h after illness onset) antiviral treatment. Although the provision of antiviral therapy between survivors and nonsurvivors did not differ significantly in our series, the importance of early treatment with antivirals in critically ill influenza patients cannot be overemphasized.
In our study, bacteremia was detected in 17 critically ill adult patients. Importantly, five of them acquired bacteremia within 48 h after hospitalization, and in three cases, the infection was caused by Staphylococcus aureus. A report of the 2009-2010 influenza pandemic among critically ill children revealed that nearly 5% of the patients had bacteremia within 72 h and Staphylococcus aureus was the most frequently isolated bacterium, which contributed to the death rate in the current pandemic [33]. In a study of 32 influenza-positive patients (including pediatric and adult patients), poor outcomes were found among patients who were coinfected with influenza viruses and Staphylococcus aureus [34]. Although we were unable to conclude whether or not initiating timely additional antimicrobial treatment in critically ill influenza patients led to better clinical outcomes, our findings and previous reports underscore that Staphylococcus aureus remains the most important cause of bacterial coinfection in pediatric and adult influenza patients.
Invasive pulmonary Aspergillus as a coinfection in patients with severe influenza has been described [35][36][37]. In a cohort study involving seven ICUs over a period of seven influenza seasons showed that influenza and the use of corticosteroids were independent risk factors for invasive aspergillosis [35]. In the present study, invasive pulmonary aspergillosis was confirmed in two deceased influenza patients with high galactomannan index in bronchoalveolar lavage fluid. This finding emphasizes that clinicians should be aware of the risk of invasive aspergillosis in critically ill influenza patients, particularly immunocompromised patients or those receiving corticosteroids. Further studies are needed to understand the incidence, risk factors, and clinical features of invasive pulmonary aspergillosis in influenza patients.
In 2003, an avian influenza virus of H5N1 subtype was isolated from a smuggled duck in Kinmen Island of Taiwan [38]. However, no locally acquired H5N1 disease in humans had been reported in Taiwan [39]. Thus, none of the patients in this series tested positive for H5N1.
This study has several potential limitations. First, given the retrospective nature of the study, information on vaccination status, including pneumococcus and influenza, and some missing laboratory data such as pneumonia severity index and coagulation profile, were not collected. Second, the study population comprised adult patients; therefore, the results cannot be generalized to pediatric patients. However, the strengths of this study include a detailed description of clinical and laboratory information at presentation and the entire hospitalization course of critically ill patients with influenza. We highlighted the key factors associated with poor outcomes for critically ill influenza patients and established a decision-making algorithm that can take advantage of simple clinical and laboratory evaluations.

Conclusions
We identified clinical and laboratory predictors of mortality that could aid in the prediction of the poor outcomes in hospitalized critically ill influenza patients, as timely intensive supportive care might be lifesaving. Medical services and ICUs can be overwhelmed during the peak of influenza epidemics, particularly in point-of-care resource-limited areas. Our findings could substantially assist with allocation of resources in the selection of the main key clinical data in primary care at the initial clinical evaluation of critically ill influenza patients.
Author Contributions: I.-K.L. made substantial contributions to the conception, study design, data analysis and interpretation, and drafting, editing, and submitting the manuscript. J.-C.H. made substantial contributions to the data collection, data analysis, and writing the manuscript. W.-C.H., Y.-C.C., and C.-Y.T. contributed to the design of the study and interpreted the findings. All authors have read and agreed to the published version of the manuscript.
Funding: This study was supported by a grant (document no. CMRPG8H0241 to I.-K.L.) from Kaohsiung Chang Gung Memorial Hospital, Kaohsiung, Taiwan. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.