A machine learning approach to identify at-risk populations for yellow fever vaccination.

Abstract

Introduction: Vaccinations play a crucial role in public and personal health. Vaccines such as the Yellow Fever 17D vaccine are effective at preventing disease and provide lifelong immunity with few adverse events. Leveraging the increase in electronic medical records and the widespread use of machine learning algorithms in healthcare, this study aims to investigate and develop an algorithmic framework for analyzing clinical data on Yellow Fever 17D vaccinations in order to predict at-risk populations for adverse events following immunizations. This research incorporates parameters from the patient's medical history and demographic information.

Methods: To build our analytic framework, we tested five machine learning algorithms: random forest, gradient boost, logistic regression, XGBoost, and Bernoulli naïve Bayes. We cleaned and managed the datasets through EHRchitect software. We assessed the performance of the algorithms using standard metrics, including precision, recall, F1-scores, accuracy, and the area under the receiver operating characteristic curve. Additionally, we implemented population sampling to mitigate potential biases common in clinical data and performed parameter ranking analysis to identify the most influential features in classifying patient reaction outcomes.

Results: We found that logistic regression produced the highest performance, and through hyperparameter optimization, we developed a framework with precision, accuracy, and ROC scores of 87.7%, 99.0%, and 0.818, respectively.

Conclusion: In conclusion, this study examined five modeling algorithms to establish a framework for analyzing real-world data and to predict at-risk populations for adverse events following immunization. We incorporated a logistic regression algorithm to assess our clinical data to achieve a high-precision performance model for classifying patient predictions.