Healthcare predictive analytics aids medical decision-making, diagnosis\nprediction and drug review analysis. Therefore, prediction accuracy is an\nimportant criteria which also necessitates robust predictive language models.\nHowever, the models using deep learning have been proven vulnerable towards\ninsignificantly perturbed input instances which are less likely to be\nmisclassified by humans. Recent efforts of generating adversaries using\nrule-based synonyms and BERT-MLMs have been witnessed in general domain, but\nthe ever increasing biomedical literature poses unique challenges. We propose\nBBAEG (Biomedical BERT-based Adversarial Example Generation), a black-box\nattack algorithm for biomedical text classification, leveraging the strengths\nof both domain-specific synonym replacement for biomedical named entities and\nBERTMLM predictions, spelling variation and number replacement. Through\nautomatic and human evaluation on two datasets, we demonstrate that BBAEG\nperforms stronger attack with better language fluency, semantic coherence as\ncompared to prior work.\n