Many people use OPUS bilingual datasets/corpora to train generic MT systems. To train specialized MT systems, you need to have in-domain sentences; in your case in pharmaceuticals. Using these specialized translated sentences, you will train (or fine-tune) an MT system to be able to translate new pharmaceuticals sentences in the future.
Medical datasets are not so popular (let alone pharmaceuticals). Plus, even if you used a publicly available dataset to train a pharmaceutical MT system, the quality is not guaranteed. What companies usually do is that they use human translators to translate their own texts for some time, and then use these translated texts to train an MT system. If you already have internal translations of your own previous documents, that what my question was about.
I hope this is clear.