Using Explainable AI Risk Predictions to Nudge Influenza Vaccine Uptake

The study team previously demonstrated that patients are more likely to receive flu vaccine after learning that they are at high risk for flu complications. Building on this past work, the present study will explore whether providing reasons that patients are considered high risk for flu complications (a) further increases the likelihood they will receive flu vaccine and (b) decreases the likelihood that they receive diagnoses of flu and/or flu-like symptoms in the ensuing flu season. It will also examine whether informing patients that their high-risk status was determined by analyzing their medical records or by an artificial intelligence (AI) / machine-learning (ML) algorithm analyzing their medical records will affect the likelihood of receiving the flu vaccine or diagnoses of flu and/or flu-like symptoms.

Geisinger has partnered with Medial EarlySign and developed an ML algorithm to identify patients at risk for serious (moderate to severe) flu-associated complications on the basis of their existing electronic health record (EHR) data. Geisinger will apply this algorithm to current patients during the 2021-22 flu season. This study will evaluate the effect of contacting patients identified as high risk with special messages to encourage vaccination. These communications will inform patients they are at high risk with either (a) no additional explanation, (b) an explanation that this determination comes from an analysis of their medical records, along with a short list of the top factors from their medical record that explain their risk, and (c) the additional explanation that an AI or ML algorithm made this determination, along with a short list of the top factors from their medical record that explain their risk. Included in the study will be current Geisinger patients 18+ years of age with no contraindications for flu vaccine and who have been assessed by the Medial algorithm and assigned a risk score. The primary study outcomes will be the rates of flu vaccination and flu diagnosis during the 2020-21 season by targeted patients.

Flu Vaccine, Choice Architecture, Machine Learning, Perceived Credibility, Personalized Risk Factors, Flu Complications

Patients from the high-risk sample will be randomly assigned to one of five arms: No-Contact Control Arm Reminder Control Arm High Risk Only Arm High Risk with Explanation Based on Medical Records Arm High Risk with Explanation Based on Algorithm Arm

Providers who prescribe vaccination and diagnose conditions will not be randomized to study arms or informed of patient assignment. Although patients will not be explicitly informed which arm they have been randomized to, they will be aware of the messages they receive.

Subjects in the no-contact control arm will receive no additional pro-vaccination intervention beyond the health system's normal efforts. Although some patients are currently targeted for flu vaccination encouragement due to a conventional non-ML assessment that they are at high risk for complications, these patients are not told that they are at high risk or that they have been targeted.

Subjects in the reminder control arm will receive messages reminding them to get the flu shot without being advised of their risk status.

Subjects in this treatment arm will receive messages telling them they have been identified to be at high risk for flu complications, without specifying how or why the health system believes this to be the case.

Subjects in this treatment arm will receive messages telling them they have been identified to be at high risk for flu complications via review of their medical records and will be provided a human-understandable short list of the top factors from their medical record that explain their risk.

Subjects in this treatment arm will receive messages telling them they have been identified to be at high risk for flu complications via analysis of their medical records by a computer algorithm and will be provided a human-understandable short list of the top factors from their medical record that explain their risk.

Received a "high confidence flu" diagnosis (with positive polymerase chain reaction [PCR]/antigen/molecular test) and/or "likely flu" diagnosis (as assessed via International Classification of Disease [ICD] codes or Tamiflu administration or positive PCR/antigen/molecular test) Note that "likely flu" is a superset of the "high confidence flu" diagnoses.

Visited ER or was hospitalized NOTE: Our prespecified outcome was description was "Visited ER, was hospitalized, or had flu-related insurance claims." We did not receive claims data, so this outcomes includes only ER visits and hospitalizations.

Inclusion Criteria: Aged 18 or older Current Geisinger patient at the time of study Falls in the top 10% of patients at highest risk, as identified by the flu-complication risk scores of machine learning algorithm (which operates on coded EHR data) Exclusion Criteria: Has contraindications for flu vaccination Has opted out of receiving communications from Geisinger via all of the modalities being tested

Data with no personally identifiable information will be made available to other researchers on the Open Science Framework for transparency. This will include the essential data and code needed to replicate the analysis that yielded reported findings.

By the paper's online publication date. Data will remain available for as long as the Open Science Framework hosts it.