I would think somewhere, someplace, maybe in a dark forbidden closet, there's someone who has access to a computer program that has the code to break the numbers and get the patient's names.
Released data isn't encoded, it's randomly labeled instead. So you don't get our infamous "Annie G Sample" being labeled as "Elpmas G Einna", or even a far more complex encoding of her name or any other data. Instead she appears as a number between 1 and 640, appearing in no particular order among the other 159 patients in her treatment arm of the trial.
The data requested would look like "CBT (trial arm), 55 (SF-36 PF starting score), 10 (CFQ bimodal starting score), 22 (CFQ likert starting score), Yes (fits ME London diagnosis)", etc. No one could conceivably be identified, or even matched with their health records, from "#10 CBT 55 10 22 Yes 60 5 12 Yes". This is the complete opposite of personal or identifiable data.
The only conceivable way that such data could identify a single patient is if 159 patients from the same treatment group got together, identified their own data, knew that Annie G Sample was in their arm of the trial, and she was not one of the people identifying her own data for the rest of the group.
But even if every one of the 159 patients in Annie G Sample's group wanted to identify her data, it's unlikely that any of them could even locate their own data based on the data requested. Can patient #52 remember if he scored 40 or 45 on his first SF-36 PF questionnaire? Or was it 50? Assuming patients were ever told their scores in the first place (they probably were not).
If I were a trial participant, I'd be far more worried about getting hit by lightning while sitting in the basement, than I'd be worried about anyone identifying me based on that data. It's simply so improbable that it's essentially impossible.