JMIR Publications

ABSTRACT

Background:

Antigenic immune escape allows seasonal influenza to persist as a global public health threat, with ongoing risk for zoonotic spillover and emergence of novel strains with pandemic potential. Emergency departments (EDs), which serve a diverse demographic of patients and regularly evaluate patients with respiratory illness, are frequently utilized for sentinel surveillance. However, existing ED-based surveillance systems are limited by the ability to identify patients whose illness severity deviates from expected patterns.

Objective:

To develop and evaluate the Expected Severity Divergence Score (ESDS), a patient-level metric designed to identify individuals whose observed clinical severity differs from model-predicted expectations, with potential application to targeted influenza surveillance.

Methods:

We conducted a retrospective cohort study across five EDs within a large academic health system (August 2018–August 2024). Adult patients with laboratory-confirmed influenza were included. Clinical, demographic, prior utilization, and medication data were extracted from the electronic health record. Two logistic regression models were developed to estimate the probability of hospital admission to either an inpatient ward or an intermediate/intensive care unit (IMC/ICU). ESDS measured the discordance between predicted admission probability and observed outcome, with higher values indicating greater divergence between expected and observed severity. Model performance was assessed using the area under the receiver operating characteristic curve (AUC). To evaluate operational feasibility, ESDS thresholds were retrospectively applied to identify cases that would have met criteria for further review, and a prototype dashboard with a simulated alerting workflow was developed.

Results:

Among 6,719 patients, 321 (4.8%) required IMC/ICU care, 1,358 (20.2%) were admitted to the inpatient floor, and 5,283 (78.6%) were discharged or transferred. Model discrimination was high for both outcomes (IMC/ICU AUC 0.814 testing; floor AUC 0.826 testing). ESDS identified subsets of patients with atypical severity profiles, where high ESDS values were associated with younger age and distinct physiologic characteristics. Retrospective application of ESDS thresholds demonstrated that a simulated alerting workflow could feasibly identify anomalous cases for potential review and targeted viral characterization.

Conclusions:

The ESDS provides a novel, patient-level approach to seasonal influenza surveillance by identifying patients whose clinical severity diverges from model-predicted expectations. Integration of the ESDE into operational surveillance may support targeted viral characterization and enhance early detection of emerging influenza variants. Prospective evaluation, external validation and assessment of clinical and public health impact are warranted.