Improving Fairness in AI Models on Electronic Health Records: The Case for Federated Learning Methods