Background:

Vaccines are a crucial tool in combating infectious diseases, yet not all vaccines are equally effective for every individual. Variations in vaccine response can be influenced by demographic factors, including age, race, gender, and ethnicity, which are sometimes overlooked in different studies. With the advancements in Artificial Intelligence (AI) and Machine Learning (ML), there is now an opportunity to predict vaccine responses more accurately. While most current models use genetic data, our study focuses on cytokine data—small proteins that affect cell interactions. Cytokine profiling measures these proteins to better understand the immune response. We initiated our investigation with two primary influenza vaccines as a pilot: the Live Attenuated Influenza Vaccine (LAIV) and the Trivalent Influenza Vaccine (TIV). This study marks the first-ever use of ML models to predict vaccine response using cytokine profiling in combination with demographic data, aiming to develop more inclusive models in healthcare.

Methods:

We analyzed cytokine profiles from over 300 participants across six influenza studies using the ImmPort database. Data were collected on days 0 (pre-vaccination), 7, and 28. Statistical analyses, including t-tests and ANOVAs, identified cytokine differences by demographics. Participants were classified as responders or non-responders using Hemagglutination Inhibition Assay (HAI) data. ML models were created using pre-vaccination cytokine profiles and demographic data to predict responses to the LAIV and the TIV. Results

Cytokine profiles significantly differed (P < 0.05) by demographics, with significant differences found in 45/55 cytokines measured. Cytokines when compared by race had the most significant differences, especially between African American and White participants, followed by age, gender, and ethnicity. For TIV, the K-Nearest Neighbors model was the most accurate for predicting responders, with accuracies of 80%, 67%, and 74% for the two A strains and one B strain. Logistic regression was most successful for LAIV with an accuracy of about 80% per strain.

Conclusion:

Our study demonstrates the value of cytokine profiling and the impact of demographics on vaccine response predictions. The integration of cytokine data with demographic factors in ML models can lead to more personalized healthcare and help address health disparities in vaccine efficacy, especially those facing African Americans. Future work will involve validating these models on larger datasets and exploring additional data types, such as flow cytometry, to enhance global health decision-making.