Predictors of Severe H1N1 Infection in Children
Predictors of Severe H1N1 Infection in Children
We identified 265 cases at 79 sites, as well as 56 additional children who fulfilled criteria for severe outcomes associated with pH1N1 but who did not have influenza-like illness at presentation (WILIE group). Boys constituted 151 (57%) cases, with a similar proportion in the two control groups ( Table 1 ). The mean age of the cases was 6.6 (SD 4.7) years; the age matched controls were of a similar age, and the random controls were an average 14 months younger. Cases were more likely to look unwell at presentation, with PRISA-II scores suggesting a need for admission to intensive care (median 19, interquartile range 14-29). No significant differences existed between cases and controls in duration of symptoms of influenza-like illness or maximum temperature recorded at home. A history of cough was present in more than 90% of cases and controls. A history of a sore throat was more prevalent in both control groups compared with cases ( Table 1 ).
Data were available for each predictor variable in at least 90% of cases and controls. Exceptions included questions concerning whether the patient had seen a physician for the presenting illness (available for 86% of age matched controls), prolonged capillary refill time (available for 69-74% in all groups), and auscultatory findings and dehydration status (available for 89% and 81% of cases, respectively) ( Table 2 and Table 3 ).
The results of the two univariable analyses comparing the cases with the random controls and the age matched controls were similar. We recorded 193 (73%) cases with pre-existing comorbidities (comparison with random controls: odds ratio 4.2, 95% confidence interval 2.8 to 6.4; comparison with age matched controls: 6.8, 4.1 to 11.1). Specifically, asthma, chronic lung disease, heart disease, preterm birth, and cerebral palsy/developmental delay were all significantly associated with case status. Furthermore, chronic renal disease (n=8, 3%) was present almost exclusively in cases. Diabetes, pregnancy, and immune suppression/malignancy occurred infrequently in both cases and controls ( Table 2 ).
At emergency department presentation, a history of dyspnoea, increased/purulent sputum, irritability, and wheezing were all significantly more common in cases than in controls, whereas diarrhoea, nausea/vomiting, generalised weakness, syncope/dizziness, myalgia, and chest pain were not. A history of seizures (n=28, 11%) and apnoea (n=10, 4%) occurred almost exclusively in cases. Headache and rhinorrhoea occurred significantly less commonly in the cases compared with both control groups. Cases were also more likely to have been previously seen by a physician for the current illness and prescribed either antivirals (exclusively oseltamivir) or antibiotics ( Table 2 ). On physical examination in the emergency department, all signs investigated, except temperature above 38.9°C, were associated with case status ( Table 3 ).
As expected, given the higher clinical severity, laboratory/radiographic investigations within four hours of arrival in the emergency department were conducted more frequently in cases than in controls; 218 (82%) of cases had chest radiography and 226 (85%) had basic laboratory investigations. In comparison, only 74 (28%) of random controls and 69 (26%) of age matched controls had chest radiography, and 45 (17%) of random controls and 44 (17%) of age matched controls had basic laboratory investigations (P<0.001 for all comparisons). In those who had laboratory and radiographic investigations within four hours of arrival in the emergency department, lobar pneumonias and platelet counts below 150 000/µL were associated with case status, whereas non-lobar pneumonias, haemoglobin concentrations below 10 g/dL, leukocyte counts above 15 000/µL, neutrophil counts above 10 000/µL, blood urea nitrogen above 20 mg/dL, and serum glucose above 200 mg/dL were not. Acidosis (pH<7.3) was exclusively present in cases, although blood pH was reported for only a few controls ( Table 3 ).
We did not include variables from laboratory and radiographic investigations in the multivariable analysis owing to the low prevalence of these variables in the control groups. The multivariable analysis using the random and the age matched controls resulted in the same six variable model that included history of chronic lung disease, history of cerebral palsy/developmental delay, requirement for oxygen or low oxygen saturations, tachycardia relative to age, presence of chest retractions, and signs of dehydration (C statistic for model using random controls 0.925; C statistic for model using age matched controls 0.905) ( Table 2 and Table 3 ). Additionally, symptoms of breathlessness, irritability/drowsiness, and increased/purulent sputum were significant at the P<0.1 level in the random control model, and symptoms of irritability/drowsiness were significant at the P<0.1 level in the age matched control model ( Supplementary Table A ). In the bootstrap analysis, we identified the same possible predictor variables at all stages of the modelling process. Sensitivity analyses, without imputation and with heart rate and respiratory rate as dichotomised variables, using the random controls resulted in a seven variable model (removal of presence of dehydration and addition of symptoms of breathlessness and irritability/drowsiness), and those using the age matched controls resulted in a three variable model (removal of a history of chronic lung disease, tachycardia relative to age, and signs of dehydration) ( Supplementary Table B ).
Of the 321 paediatric patients with severe pH1N1 outcome identified, 265 (83%) fulfilled the Centers for Disease Control and Prevention criteria for influenza-like illness at emergency department presentation (cases). The additional 56 patients in the WILIE group had similar demographics and comorbidities to cases. However, more children in the WILIE group had immune suppression/malignancy (n=4, 7%; influenza-like illness cases n=5, 2%; odds ratio 4.0, 1.0 to 5.4).
Within four hours of arrival at the emergency department, 63% of both cases and WILIE patients received antibiotic treatment, 40% received antiviral treatment, 64% received intravenous fluid boluses, 42% had ventilatory support started/continued, and 14% had inotropic support started. Two hundred and eighty six (89%) patients were admitted to hospital at the time of their first emergency department presentation. Most admitted patients subsequently received antibiotic and antiviral treatment. Approximately one half of the children received systemic corticosteroid treatment, and a small number received immunoglobulin treatment. The most common complications during hospital admissions were secondary pneumonias, secondary bacteraemias, and acute respiratory distress syndrome ( Table 4 ).
Among the 321 patients with severe pH1N1, 34 (11%) deaths occurred (27 (10%) deaths among cases and 7 (13%) among WILIE patients). Two patients arrived dead at the emergency department. The other 32 who subsequently died were less likely to have received antiviral treatment within the first four hours of initial emergency department presentation than were those who survived (P=0.02). Those who died received greater inotropic and ventilatory support and developed more frequent complications ( Table 5 ). Of the seven variables identified in the multivariable analyses, only cerebral palsy/developmental delay was associated with an increased risk of death (odds ratio 2.6, 1.2 to 5.5).
Results
We identified 265 cases at 79 sites, as well as 56 additional children who fulfilled criteria for severe outcomes associated with pH1N1 but who did not have influenza-like illness at presentation (WILIE group). Boys constituted 151 (57%) cases, with a similar proportion in the two control groups ( Table 1 ). The mean age of the cases was 6.6 (SD 4.7) years; the age matched controls were of a similar age, and the random controls were an average 14 months younger. Cases were more likely to look unwell at presentation, with PRISA-II scores suggesting a need for admission to intensive care (median 19, interquartile range 14-29). No significant differences existed between cases and controls in duration of symptoms of influenza-like illness or maximum temperature recorded at home. A history of cough was present in more than 90% of cases and controls. A history of a sore throat was more prevalent in both control groups compared with cases ( Table 1 ).
Predictor Variable Data Availability
Data were available for each predictor variable in at least 90% of cases and controls. Exceptions included questions concerning whether the patient had seen a physician for the presenting illness (available for 86% of age matched controls), prolonged capillary refill time (available for 69-74% in all groups), and auscultatory findings and dehydration status (available for 89% and 81% of cases, respectively) ( Table 2 and Table 3 ).
Univariable Analysis
The results of the two univariable analyses comparing the cases with the random controls and the age matched controls were similar. We recorded 193 (73%) cases with pre-existing comorbidities (comparison with random controls: odds ratio 4.2, 95% confidence interval 2.8 to 6.4; comparison with age matched controls: 6.8, 4.1 to 11.1). Specifically, asthma, chronic lung disease, heart disease, preterm birth, and cerebral palsy/developmental delay were all significantly associated with case status. Furthermore, chronic renal disease (n=8, 3%) was present almost exclusively in cases. Diabetes, pregnancy, and immune suppression/malignancy occurred infrequently in both cases and controls ( Table 2 ).
At emergency department presentation, a history of dyspnoea, increased/purulent sputum, irritability, and wheezing were all significantly more common in cases than in controls, whereas diarrhoea, nausea/vomiting, generalised weakness, syncope/dizziness, myalgia, and chest pain were not. A history of seizures (n=28, 11%) and apnoea (n=10, 4%) occurred almost exclusively in cases. Headache and rhinorrhoea occurred significantly less commonly in the cases compared with both control groups. Cases were also more likely to have been previously seen by a physician for the current illness and prescribed either antivirals (exclusively oseltamivir) or antibiotics ( Table 2 ). On physical examination in the emergency department, all signs investigated, except temperature above 38.9°C, were associated with case status ( Table 3 ).
As expected, given the higher clinical severity, laboratory/radiographic investigations within four hours of arrival in the emergency department were conducted more frequently in cases than in controls; 218 (82%) of cases had chest radiography and 226 (85%) had basic laboratory investigations. In comparison, only 74 (28%) of random controls and 69 (26%) of age matched controls had chest radiography, and 45 (17%) of random controls and 44 (17%) of age matched controls had basic laboratory investigations (P<0.001 for all comparisons). In those who had laboratory and radiographic investigations within four hours of arrival in the emergency department, lobar pneumonias and platelet counts below 150 000/µL were associated with case status, whereas non-lobar pneumonias, haemoglobin concentrations below 10 g/dL, leukocyte counts above 15 000/µL, neutrophil counts above 10 000/µL, blood urea nitrogen above 20 mg/dL, and serum glucose above 200 mg/dL were not. Acidosis (pH<7.3) was exclusively present in cases, although blood pH was reported for only a few controls ( Table 3 ).
Multivariable Models
We did not include variables from laboratory and radiographic investigations in the multivariable analysis owing to the low prevalence of these variables in the control groups. The multivariable analysis using the random and the age matched controls resulted in the same six variable model that included history of chronic lung disease, history of cerebral palsy/developmental delay, requirement for oxygen or low oxygen saturations, tachycardia relative to age, presence of chest retractions, and signs of dehydration (C statistic for model using random controls 0.925; C statistic for model using age matched controls 0.905) ( Table 2 and Table 3 ). Additionally, symptoms of breathlessness, irritability/drowsiness, and increased/purulent sputum were significant at the P<0.1 level in the random control model, and symptoms of irritability/drowsiness were significant at the P<0.1 level in the age matched control model ( Supplementary Table A ). In the bootstrap analysis, we identified the same possible predictor variables at all stages of the modelling process. Sensitivity analyses, without imputation and with heart rate and respiratory rate as dichotomised variables, using the random controls resulted in a seven variable model (removal of presence of dehydration and addition of symptoms of breathlessness and irritability/drowsiness), and those using the age matched controls resulted in a three variable model (removal of a history of chronic lung disease, tachycardia relative to age, and signs of dehydration) ( Supplementary Table B ).
Outcome for Cases and the WILIE Group
Of the 321 paediatric patients with severe pH1N1 outcome identified, 265 (83%) fulfilled the Centers for Disease Control and Prevention criteria for influenza-like illness at emergency department presentation (cases). The additional 56 patients in the WILIE group had similar demographics and comorbidities to cases. However, more children in the WILIE group had immune suppression/malignancy (n=4, 7%; influenza-like illness cases n=5, 2%; odds ratio 4.0, 1.0 to 5.4).
Within four hours of arrival at the emergency department, 63% of both cases and WILIE patients received antibiotic treatment, 40% received antiviral treatment, 64% received intravenous fluid boluses, 42% had ventilatory support started/continued, and 14% had inotropic support started. Two hundred and eighty six (89%) patients were admitted to hospital at the time of their first emergency department presentation. Most admitted patients subsequently received antibiotic and antiviral treatment. Approximately one half of the children received systemic corticosteroid treatment, and a small number received immunoglobulin treatment. The most common complications during hospital admissions were secondary pneumonias, secondary bacteraemias, and acute respiratory distress syndrome ( Table 4 ).
Mortality
Among the 321 patients with severe pH1N1, 34 (11%) deaths occurred (27 (10%) deaths among cases and 7 (13%) among WILIE patients). Two patients arrived dead at the emergency department. The other 32 who subsequently died were less likely to have received antiviral treatment within the first four hours of initial emergency department presentation than were those who survived (P=0.02). Those who died received greater inotropic and ventilatory support and developed more frequent complications ( Table 5 ). Of the seven variables identified in the multivariable analyses, only cerebral palsy/developmental delay was associated with an increased risk of death (odds ratio 2.6, 1.2 to 5.5).