Can I retrain the model with samples who have low OOB error?

Question

I trained a RF regression model for blood pressure estimation from PPG signal and tested it with Out-Of-Bag (OOB) error. I find that for several samples from my dataset (about 10% of dataset) the OOB prediction is poor. Is it possible to separate the samples with poor OOB prediction and retrain the model with samples that have good OOB prediction? Is this a good idea for improving the model accuracy? Thanks a lot