International Classification of Diseases (ICD) codes obtained from real-world data can be used to identify influenza cases for epidemiological research but, without validation, may introduce biases. The objective of this study was to validate ICD influenza discharge diagnoses using real-time reverse transcription-polymerase chain reaction (RT-PCR) laboratory-confirmed influenza (LCI) results. The study was conducted during six influenza seasons (2012/2013-2017/2018) in the Valencia Hospital Surveillance Network for the Study of Influenza (VAHNSI). Patients aged 18+ years were identified via active-surveillance and had to meet an influenza-like illness (ILI) case definition to be included. All patients were tested for influenza by real-time RT-PCR. Main and secondary influenza discharge diagnosis codes were extracted from hospital discharge letters. Positive predictive values (PPVs) and the complementary of the sensitivities (1-Sensitivity) of ICD codes with corresponding 95% credible intervals (CrIs) were estimated via binomial Bayesian regression models. A total of 13,545 patients were included, with 2257 (17%) positive for influenza. Of 2257 LCI cases, 1385 (61%) were not ICD-coded as influenza. Overall, 74.73% (95% CrI: 63.24-84.44) of LCI were not-ICD coded as influenza (1-Sensitivity) after adjustment. Sensitivity improved across seasons and with increasing age. Average PPV was 74.02% (95% CrI: 68.58-79.17), ranging from 43.71% to 81.57% between seasons. Using only main and secondary discharge diagnosis codes for influenza detection markedly underestimates the full burden of influenza in hospitalized patients. Future studies, including post-COVID context, using prospective surveillance for ILI are required to assess the validity of hospital discharge data as a tool for determining influenza-related burden of disease.
Read full abstract