Rule-Based High-Risk Flu

Overview

Previous work by the study team has shown that informing patients of their high risk of flu and flu-related complications increases their likelihood of getting a flu shot (Rosenbaum et al., 2026). In this work, an artificial intelligence (AI)-based algorithm determined which patients were at high risk for flu. This year, the team will test whether risk messages remain effective when a non-algorithmic rule-based additive risk index (McGovern et al., 2024) is used to identify patients at high risk.

While most individuals recover from influenza without complications, certain populations-including older adults and those with underlying medical conditions-are at increased risk for severe outcomes such as pneumonia, other respiratory complications, and death. Identifying these high-risk patients is therefore important for targeting outreach and improving vaccination uptake. Over the past three influenza seasons, Geisinger has sent as standard of care messages to patients identified as high risk for flu and flu-related complications, informing them of their risk and encouraging vaccination. These messages were implemented following a series of four randomized controlled trials conducted by the study team, which demonstrated that such messages increased influenza vaccination rates (Rosenbaum et al., 2026). Previously, high-risk patients were identified using an AI-based model. Patients were considered high risk if they were in the top 15% of risk among Geisinger patients eligible for scoring by the model. Due to logistical and resource constraints, this model is no longer available for use. As a result, an alternative, scalable approach is needed to identify high-risk patients for upcoming influenza seasons. Recent evidence suggests that a simple count of Center for Disease Control (CDC)-defined influenza risk factors is predictive of severe influenza outcomes (McGovern et al., 2024). This approach constructs a rule-based additive risk index by summing the number of CDC-defined high-risk conditions present in each patient's electronic health record. Results indicate that this simple index is highly informative for identifying patients at increased risk of influenza-related complications. This study will evaluate whether a rule-based additive risk index derived from Electronic Health Record (EHR) data can be used to identify patients at high risk for influenza and influenza-related complications and support targeted outreach. Specifically, the study will assess whether informing patients with four or more CDC risk factors (representing approximately the top 15% of the distribution of scores for eligible patients) of their elevated risk increases influenza vaccination uptake. Of the 51218 patients who met inclusion criteria, 8750 (17.1%) were randomized to the control group to allow for 80% power to detect a 1.4 percentage-point difference in vaccination rate between control and high-risk outreach groups, with a baseline of 23% and 2-tailed p\<.05. Outreach in the high-risk outreach group will be completed by the Marketing team, using filters for each study modality (mailed letter, patient portal message, text message) including filters for patient contact preferences for patients in the outreach group. In addition to a letter if eligible, patients will be sent either a portal message or a text message (not both) based on their preferences. Marketing cannot apply the same filters to patients in the control group, so to keep analyses unbiased, the primary analyses will be intent-to-treat. If possible, the study team will attempt to rebuild the filters applied by the Marketing team, and apply them consistently to both high-risk outreach and control groups for a focused exploratory analysis among patients eligible for the messages.

Conditions

Interventions

Eligibility

Locations (1)

Outcomes

Central Contacts