Data from: Clinical and laboratory predictors of influenza infection among individuals with influenza-like illness presenting to an urban Thai hospital over a five-year period

Early diagnosis of influenza infection maximizes the effectiveness of
antiviral medicines. Here, we assess the ability for clinical
characteristics and rapid influenza tests to predict PCR-confirmed
influenza infection in a sentinel, cross-sectional study for
influenza-like illness (ILI) in Thailand. Participants meeting criteria
for acute ILI (fever > 38°C and cough or sore throat) were
recruited from inpatient and outpatient departments in Bangkok, Thailand,
from 2009-2014. The primary endpoint for the study was the occurrence of
virologically-confirmed influenza infection (based upon detection of viral
RNA by RT-PCR) among individuals presenting for care with ILI. Nasal and
throat swabs were tested by rapid influenza test (QuickVue) and by RT-PCR.
Vaccine effectiveness (VE) was calculated using the case test-negative
method. Classification and Regression Tree (CART) analysis was used to
predict influenza RT-PCR positivity based upon symptoms reported. We
enrolled 4572 individuals with ILI; 32.7% had detectable influenza RNA by
RT-PCR. Influenza cases were attributable to influenza B (38.6%),
A(H1N1)pdm09 (35.1%), and A(H3N2) (26.3%) viruses. VE was highest against
influenza A(H1N1)pdm09 virus and among adults. The most important symptoms
for predicting influenza PCR-positivity among patients with ILI were
cough, runny nose, chills, and body aches. The accuracy of the CART
predictive model was 72.8%, with an NPV of 78.1% and a PPV of 59.7%.
During epidemic periods, PPV improved to 68.5%. The PPV of the QuickVue
assay relative to RT-PCR was 93.0% overall, with peak performance during
epidemic periods and in the absence of oseltamivir treatment. Clinical
criteria demonstrated poor predictive capability outside of epidemic
periods while rapid tests were reasonably accurate and may provide an
acceptable alternative to RT-PCR testing in resource-limited areas.