Identification and characterisation of clinically distinct subgroups of adults hospitalised with influenza in the USA: a repeated cross-sectional study

Abstract

Background: Patients hospitalised with influenza have heterogeneous clinical presentations and disease severity, which may complicate epidemiologic study design or interpretation. We applied latent class analysis to identify clinically distinct subgroups of adults hospitalised with influenza.

Methods: We analysed cross-sectional study data on adults (≥18 years) hospitalised with laboratory-confirmed influenza from the population-based U.S. Influenza Hospitalization Surveillance Network (FluSurv-NET) including 13 states during 2017-2018 and 2018-2019 influenza seasons (October 1 through April 30). Adults were included if they were residents of the FluSurv-NET catchment area, hospitalised with laboratory-confirmed influenza during these two seasons, and had both the main case report form and the supplemental disease severity case report form completed. We constructed a latent class model to identify subgroups from multiple observed variables including baseline characteristics (age and comorbidities) and clinical course (symptoms at admission, respiratory support requirement, and development of new complications and exacerbations of underlying conditions).

Findings: Among the 43,811 influenza-associated hospitalizations reported during the 2017-2018 and 2018-2019 influenza seasons, 15,873 (36.2%) were included in our analytic population: among them, 7069 (44.5%) were male and 8804 (55.5%) were female. We identified five subgroups. Subgroup A included persons of all ages with few comorbidities and 87.9% (255/290) of pregnant women. Subgroup B included older adults with comorbidities (cardiovascular disease (79.7% [3650/4581]) and diabetes (50.6% [2320/4581])). Almost all patients in subgroups C and D had asthma or chronic lung disease and high proportions with exacerbations of underlying conditions (59.7% [889/1489] and 65.1% [2274/3496], respectively). Subgroup E had the highest proportion with new complications (90.3% [1383/1531]). Subgroups D and E had the highest proportions with severe disease indicators: 21.0% (733/3496) and 50.4% (771/1531) required ICU admission, 7.2% (253/3496) and 28.0% (428/1531) required invasive mechanical ventilation, and 3.3% (116/3496) and 11.4% (174/1531) died in-hospital, respectively.

Interpretation: The five identified subgroups of adults hospitalised with influenza had varying distributions of age, comorbid conditions, and clinical courses characterized by new complications versus exacerbations of existing conditions. Stratifying by these subgroups may strengthen analyses that assess the impact of influenza vaccination and antiviral treatment on risk of severe disease. Limitations included that results were based on a convenience sample within FluSurv-NET sites and were likely not representative of all adults hospitalised with influenza in the United States. Influenza testing was also clinician-driven, likely leading to under-ascertainment.

Funding: Centers for Disease Control and Prevention.

Keywords: Characterization; Disease severity; Hospitalization; Influenza; Latent class analysis.

Fig. 1

Flow chart of inclusion and exclusion criteria for the analytic population.

Fig. 2

Schematic showing the variables included in the latent class model and descriptions of the five identified subgroups.

Fig. 3

Prevalence of each indicator of disease severity within subgroups of adults hospitalised with influenza.