Innate Immune Function and Mortality in Critically Ill Children With Influenza: A Multicenter Study^*

Patients

Patients ≤18 yrs of age admitted to a PICU with community-acquired influenza infection from December 2008 to November 2009 were eligible to participate. Influenza infection was defined as a positive test for influenza by direct antigen test, direct fluorescent antibody test, polymerase chain reaction test, or viral culture. Additional inclusion criteria included the presence of 1) evidence of lower respiratory tract infection; 2) acute respiratory failure treated with mechanical ventilation regardless of etiology; 3) central nervous system involvement (e.g., coma, seizure); or 4) shock requiring vasopressors. Lower respiratory tract infection was defined as having any of the following: 1) an infiltrate or effusion on chest radiograph and need for supplemental oxygen to maintain systemic oxygen saturations ≥95%; 2) the presence of rales, wheezes, or decreased aeration on chest auscultation; or 3) tachypnea, retractions, or shortness of breath. Shock was defined as receiving >60 mL/kg of resuscitative fluid and/or inotrope or vasopressor support. Exclusion criteria included the presence of isolated upper respiratory tract infection in the absence of other inclusion criteria and/or the presence of nosocomially acquired influenza infection as defined by the onset of influenza-like symptoms and a positive influenza test >48 hrs from hospital admission. After July 7, 2009, we began excluding patients with underlying conditions (respiratory, neurologic, cardiovascular, immunologic, metabolic, and endocrine) predisposing to more severe influenza infection as defined by the Advisory Committee on Immunization Practices (22) and patients who were not severely critically ill (those without marked impairment in mental status or without use of vasopressors or mechanical ventilator support). These steps were taken to enrich our cohort with previously healthy patients who had severe critical illness, a subset of patients that was underrepresented in our initial cohort.

Throughout the entire study period, sites followed the recommendations published by the Centers for Disease Control and Prevention, which called for screening all symptomatic patients admitted to the PICU for influenza and testing all intubated patients for secondary bacterial infection by sending a Gram stain and culture of endotracheal secretions (CD-CHAN-00268-2008-01-30-ADV-N). Patient management was at the discretion of the treating clinicians.

Contemporaneous outpatient control subjects were recruited from the outpatient phlebotomy area at Nationwide Children’s Hospital. Children aged 0–18 yrs presenting for elective phlebotomy were eligible. Children were excluded if they had subjective or objective fever within the past 24 hrs, use of nonsteroidal anti-inflammatory drugs within the past 48 hrs, pre-existing immunosuppression including history of transplantation of any kind, active treatment for malignancy, active treatment of an autoimmune disorder, systemic corticosteroid treatment within the past month, and presence of a known immunosuppressive medication on the patient’s current medication list.

Laboratory Testing

Enrolled influenza subjects included in these analyses underwent blood sampling at a single time point within 72 hrs of ICU admission. Intubated subjects also underwent endotracheal tube sampling for viral coinfection screening using FilmArray multiplex polymerase chain reaction (Idaho Technology, Salt Lake City, UT).

For serum cytokine analyses, whole blood was collected in an anticoagulant-free tube (Becton Dickinson, Franklin Lakes, NJ) and allowed to clot for 30 mins. Tubes were then centrifuged at 1000 × g for 10 mins (within 60 mins of collection), and serum was aliquoted and stored at −80°C and sent to the Cytokine Reference Laboratory at the University of Minnesota where cytokine/chemokine levels were quantified by multiplex assay on the Luminex platform (Luminex, Austin, TX) and Bioplex 2.0 software (Biorad, Hercules, CA). Cytokines and chemokines evaluated with a human-specific 30-plex (Millipore, Billerica, MA) included TNF-α, interferon (IFN)-α2, IFN-γ, eotaxin (CCL11), granulocyte colony stimulating factor, granulocyte-macrophage colony stimulating factor (GM-CSF), growth-related oncogene-α (CXCL1), interleukin (IL)-1α, IL-1β, IL-1 receptor antagonist, IL-2, IL-4, IL-5, IL-6, IL-7, IL-8, IL-10, IL-12 p40, IL-12 p70, IL-13, IL-17, IFN-inducible protein (IP)-10 (CXCL10), monocyte chemotactic protein (MCP)-1 (CCL2), macrophage-derived chemokine (CCL22), macrophage inflammatory protein (MIP)-1α (CCL3), MIP-1β (CCL4), regulated on activation normal T cell expressed and secreted (CCL5), soluble CD 40 ligand (sCD40L), vascular endothelial growth factor, and fractalkine. IFN-β was evaluated as a single plex using a bead set from Panomics (Fremont, CA). All values were interpolated from standard curves of the corresponding human recombinant proteins provided with the multiplex kits.

Quantification of the capacity of subjects’ innate immune system to respond to a challenge was done by measurement of ex vivo LPS-induced TNF-α production capacity. Whole blood was collected in a heparinized tube (Becton Dickinson, Franklin Lakes, NJ) and placed on ice. Within 60 mins of collection, 50 μL of whole blood was added to a tube containing 500 pg/mL LPS (phenol-extracted from Salmonella abortus equi [Sigma, St. Louis, MO]) and incubated for 4 hrs at 37°C. This was done in duplicate for each sample. Tubes were then centrifuged (1000 × g for 5 mins), and supernatants were frozen at −80°C. To assure uniformity of results, LPS stimulation solution was made and shipped monthly by the Immune Surveillance Laboratory at The Research Institute at Nationwide Children’s Hospital. Kits were quality controlled at the site of manufacture such that intrabatch variation in TNF-α stimulation capacity was <10%. Supernatants from each participating study site were shipped quarterly to the Immune Surveillance Laboratory for TNF-α determination. TNF-α was quantified in post-stimulation supernatants using the Immulite automated chemiluminometer (Siemens Healthcare Diagnostics, Deerfield, IL), and average values from duplicate samples were reported.

Although complete blood counts were not part of the study protocol, complete blood counts and differential data obtained within 24 hrs of immune function testing were used to calculate absolute cell counts when available.

Outpatient control subjects also underwent measurement of whole blood ex vivo TNF-α production capacity as well as quantitation of a more limited panel of plasma cytokines (IL-6 [detection threshold: 2 pg/mL] and IL-10 [detection threshold: 5 pg/mL]) using the Immulite automated chemiluminometer. These cytokines were chosen to screen for evidence of systemic inflammation and the counterregulatory anti-inflammatory response in these subjects.

Definitions

Race and ethnicity (Hispanic or non-Hispanic) were defined by the child’s parents. Pediatric Risk of Mortality III acute physiology score (PRISM III) (23) was used to measure severity of illness within the first 24 hrs of admission. Noninvasive ventilation was defined as delivery of continuous positive airway pressure or bilevel positive airway pressure from a mechanical ventilator through a mask. Acute lung injury and acute respiratory distress syndrome (ARDS) were defined as acute onset of hypoxia (Pao₂/Fio₂ ratio of ≤300 for acute lung injury and ≤200 for ARDS) with bilateral infiltrates on chest radiograph and no evidence of left heart failure. Shock requiring inotropic or vasopressor support was defined as the use of a dopamine infusion >5 μg/kg/min or any epinephrine, norepinephrine, phenylephrine, dobutamine, or milrinone infusion to maintain blood pressure. Viral coinfections were defined as polymerase chain reaction positivity for any virus other than influenza from study airway samples using the FilmArray method. Bacterial secondary infections were defined as the growth of any pathogen from a sterile body site (lower airway, blood, urine) or polymerase chain reaction positivity for known pathogens (e.g., Mycoplasma species) from samples obtained within 48 hrs of ICU admission. Mortality was defined as death during the hospitalization.

Statistical Analyses

The clinical and cytokine markers were evaluated and summarized both quantitatively and graphically through exploratory data analyses. Between-group comparisons were made with the Mann–Whitney U test for continuous variables and Fisher’s exact test or χ² test for categorical variables. Univariate and multivariable logistic regression models were created to evaluate the relationships among ex vivo TNF-α production capacity, clinical covariables, and mortality. Covariables for the multivariate models included PRISM III score, presence of secondary bacterial infection on admission, presence of Staphylococcus aureus infection on admission, 2009 H1N1 status, presence of at least one underlying condition, and presence of an immunocompromised state at baseline. Identification of these clinical factors to include was based on univariate modeling results as well as all subsets regression. Given the limited numbers of nonsurvivors, multivariable analyses were restricted to two covariates (ex vivo TNF-α production capacity and a clinical factor). To evaluate goodness of fit in the logistic regression models, we looked at the model’s likelihood ratio tests as well as the (pseudo) R² tests that were specific to the logistic regression setting. Cytokine values (including ex vivo TNF-α production capacity) were log-transformed for regression analyses. A receiver operating characteristic curve with area under the curve analysis was used to further evaluate the relationship between ex vivo LPS-induced TNF-α production capacity and mortality.

The number of ICU-free days was also evaluated using univariate and multivariable generalized linear regression models in a similar approach as that described for the hospital mortality analyses but without the event-based constraints. Finally, time to ICU discharge was evaluated, where the number of days in the ICU until discharge in the first 28 days was used; those patients who were still in the ICU at 28 days were censored at 28 days. Those patients who died before 28 days in the ICU were treated as competing events. Cumulative incidence models were used to assess the time to ICU discharge while controlling for the competing risk of hospital mortality within 28 days.

A p value < 0.05 was considered significant throughout, except for the serum cytokine/chemokine analyses where a p value of < 0.0016 was considered significant after Bonferroni correction for multiple comparisons within the 31-mediator panel (0.05/31). Analyses were done using the Windows version of the statistical program R Version 2.11.1 (The R Foundation for Statistical Computing, Vienna, Austria) and Prism5 (GraphPad, La Jolla, CA) software.

Subjects

Of 197 eligible subjects, 146 (73%) were enrolled. Of these, 53 (36%) underwent sampling within the first 72 hrs of PICU admission. One subject (a survivor) was excluded from analysis as a result of deviation from the specimen collection protocol, leaving 52 (44 survivors [85%], eight nonsurvivors [15%]). In the cohort as a whole, survivors and nonsurvivors were similar in terms of gender, age, ethnicity, race, and presence of comorbidities (Table 1).

The ICU clinical characteristics of influenza subjects are shown in Table 2. The median time between ICU admission and blood sampling was in the second ICU day for both survivors and nonsurvivors. All nonsurvivors were invasively mechanically ventilated compared with 68% of survivors. Among those receiving invasive mechanical ventilation, high-frequency oscillatory ventilation was used in 21% of patients (eight of 38). Of these, two were survivors and six were nonsurvivors. The exclusive use of noninvasive ventilatory support was rare, accounting for only 5.8% of subjects (all survivors). Extracorporeal membrane oxygenation (ECMO) support was used in six subjects, all nonsurvivors, four of whom were on ECMO at the time of sample collection. Nonsurvivors had higher PRISM III scores, were more likely to have shock (eight of eight) and ARDS (seven of eight), and demonstrated a trend toward longer lengths of ICU stay. Forty-one subjects (79%) had a complete blood count with differential performed within 24 hrs of study sampling (33 survivors, eight nonsurvivors). In this subset, nonsurvivors were significantly more leukopenic with the drop in cell numbers appearing to be consistent across the monocyte, neutrophil, and lymphocyte lineages.

The microbiologic characteristics of influenza subjects are shown in Table 3. Nearly half of subjects (48%) were confirmed or suspected to have had 2009 H1N1 with two-thirds of the remaining subjects positive for seasonal influenza A. Children with 2009 H1N1 were more likely to have had shock (p = 0.02), ARDS (p = 0.005), and S. aureus secondary bacterial infection (p = 0.02). The 2009 H1N1 was more prevalent in nonsurvivors (seven of eight) than survivors (18 of 44) and was significantly associated with mortality (p = 0.02). There was no difference in the use of antiviral medications either during the hospital course or within 48 hrs of symptom onset between survivors and nonsurvivors.

Secondary bacterial infections were diagnosed at admission more frequently in nonsurvivors (six of eight) than survivors (12 of 44). The airway was the primary source of secondary bacterial infection in both nonsurvivors (six of six) and survivors (11 of 12) with one survivor demonstrating bacteremia. Having any secondary bacterial infection at admission was significantly associated with mortality (p = 0.015), but that association was stronger if the infection with S. aureus (p = 0.007). Among nonsurvivors with secondary bacterial infections, S. aureus was the most common pathogen (five of six) with one nonsurvivor having pneumonia with Enterobacter cloacae. S. aureus also was the most common isolate among survivors with secondary bacterial infection (six of 12), with group A streptococcus, Haemophilus influenza, Moraxella catarrhalis, Mycoplasma pneumoniae, and Pseudomonas aeruginosa also represented. Notably, all of the S. aureus isolates from nonsurvivors were methicillin-resistant (MRSA), whereas only one of the six staphylococcal isolates from survivors was MRSA. Additional viruses were identified commonly in the respiratory secretions of both survivors (39%) and nonsurvivors (25%), with rhinovirus being the most common among all subjects (eight of 52).

Control Subjects

Twenty-one outpatients were enrolled and sampled at Nationwide Children’s Hospital as control subjects. They included 11 males and ten females with a median age of 11 yrs (range, 8.2–13 yrs) (p = 0.14 vs. all influenza subjects, p = 0.84 vs. influenza nonsurvivors). Influenza testing was not performed on these subjects, but all were afebrile by history and had no influenza-like symptoms at the time of sampling.

Systemic Cytokine/Chemokine Levels

As shown in Table 4, serum cytokine/chemokine profiles were significantly different between survivors and nonsurvivors. Six mediators (GM-CSF, IL-6, IL-8, IP-10, MCP-1, and MIP-1α) were significantly higher in nonsurvivors at an α level < 0.0016, thus meeting the Bonferroni correction for multiple comparisons (see Supplement Digital Content 1, http://links.lww.com/CCM/A519 for results from the full panel of mediators). In contrast, systemic cytokine concentrations were low among outpatient controls with IL-6 and IL-10 levels typically at or near the threshold of detection: IL-6: 2.6 (2.0–5.8) pg/mL and IL-10: 5 (5) pg/mL.

Children with 2009 H1N1 had significantly higher serum levels of IFN-γ, GM-CSF, IL-8, IL-17, MIP-1α, and vascular endothelial growth factor compared with children with other influenza types. The median concentrations of these mediators, however, were typically quite low even in children with 2009 H1N1 (see Supplement Digital Content 2, http://links.lww.com/CCM/A519). Children with S. aureus secondary infection had higher levels of IFN-γ, GM-CSF, IL-8, IL-17, MIP-1α, and vascular endothelial growth factor compared with children with either no secondary infection or infection with a different organism. In these analyses, the median concentrations of mediators were higher in the S. aureus group than those seen in the 2009 H1N1 analyses (see Supplement Digital Content 3, http://links.lww.com/CCM/A519).

Ex Vivo TNF-α Production Capacity

Despite high levels of circulating proinflammatory cytokines, critically ill children with influenza demonstrated lower ex vivo TNF-α production capacity compared with outpatient control subjects (610 [254–934] vs. 1297 [1088–1739] pg/mL, p < 0.0001) (Fig. 1A). Furthermore, influenza nonsurvivors had markedly lower ex vivo TNF-α production capacity compared with survivors (157 [138–206] vs. 667 [364–997] pg/mL, p < 0.0001) (Fig. 1B). Reduced ex vivo TNF-α production capacity was associated with mortality by univariate logistic regression (p = 0.009) and remained significantly associated with mortality after controlling for PRISM III score, presence of secondary bacterial infection at admission, 2009 H1N1 status, presence of underlying conditions, and baseline immunocompromise (Table 5). Receiver operating characteristic curve analysis indicates an excellent ability of the ex vivo TNF-α production capacity to distinguish between survivors and nonsurvivors (area under the curve: 0.97, p < 0.0001) with an optimum discriminatory cutoff of 249 pg/mL (Fig. 2A).

On analyzing ICU-free days through day 28 (when those who died were given a value of 0), lower ex vivo TNF-α production capacity was associated with fewer ICU-free days by univariate linear regression (p = 0.009). The clinical variable that was most strongly associated with length of ICU stay by linear regression was PRISM III score (p = 0.0005). When performing bivariate regression including PRISM III score, ex vivo TNF-α production capacity remains significantly associated with ICU-free days (p = 0.04). Similarly, by time-to-event analysis, with death as a competing factor, ex vivo TNF-α production is significantly associated with time to ICU discharge (p = 0.005). If ex vivo TNF-α production is dichotomized (<250 pg/mL vs. ≥250 pg/mL), this relationship becomes stronger (p = 0.0006) (Fig. 2B). This remains significant after adjusting for PRISM III score (p = 0.01).

Among all subjects with influenza, children with 2009 H1N1 had lower ex vivo TNF-α production capacity than children with other influenza strains (256 [167–1060] vs. 696 [398–906], p = 0.046). Similarly, children with S. aureus secondary infection had lower TNF-α responses than children with either no bacterial coinfection or coinfection with different bacteria (Fig. 1C). There was no difference in ex vivo TNF-α production capacity between children with and without underlying medical conditions (470 [176–1025] vs. 623 [300–906], p = 0.50).

There was an association between lower absolute monocyte counts (AMC) in children with lower ex vivo TNF-α production capacity (p = 0.002, R² = 0.23) (Fig. 2C). The highest AMC among nonsurvivors was 282 cells/mm³. There were nine survivors with AMC at or below this level and eight of nine of those had an ex vivo TNF-α production capacity >250 pg/mL. Although both thresholds (AMC ≤288 cells/mm³ and ex vivo TNF-α production capacity <250 pg/mL) demonstrated 100% specificity and positive predictive value for mortality, ex vivo TNF-α production was more sensitive (67% vs. 47%) with better negative predictive value (91% vs. 73%).