DMcG, I am intrigued by your post.

"the way drugs are developed and tested is strongly focused on the their effects in isolation". Well yes (if we discount, for example, adjuvant and neoadjuvant trials). The purpose of a trial is to discover if a drug is (1) safe and (2) more effective than anything else currently on the market for that particular condition.

"You ultimately rely on an individual patient who is taking two or more drugs to report symptoms, that the doctor concerned notices a connection and reports that upwards until eventually that is assessed and recorded as an incompatibility". I don't think so. I think you are saying that "incompatibilities" - side effects? - are "ultimately" discovered at the micro - or individual - level? My experience is that adverse events are examined at a macro - or large dataset (not necessarily so-called big data) level, where the confidence intervals and statistical powers can be calculated and hypotheses tested (not by physicians, but by statisticians and/or epidemiologists - most physicians of my acquaintance delight in saying how frightened they are of statistics). These data are collected from anonymised GP records, hospital data or (in the case of, e.g., the US) medical insurance information. As a freelancer, every contract I have signed with a pharma client in the past twenty years has included an agreement to report - immediately - any adverse event occurring to someone taking one of their drugs (be it someone at a party telling me about their uncle, or even hearing an item on a radio programme) - information that is added to these data above. Also, I'm not sure that, even if you sign yourself out of GPDR, data concerning you that has been anonymised will be excluded from datasets created for the purpose of pharmacovigilance. If these data cannot be linked to you then they might not be considered personal data.

"Detecting of subtle relationships and whatnot is almost entirely in the commercial sector with things like Palantir". Drugs are mostly developed in the commercial sector, so unsurprisingly their long-term effects are also calculated almost entirely in the commercial sector, although this can be at the behest of the regulatory authorities as statutory requirements. This may not be the best way, but it's how things work at the moment. Universities and hospitals also do this kind of research, sometimes in partnership with pharma companies. There are plenty of statistical methods to detect interactions, relationships, correlations, clusters, etc., so I'm not sure why you mention Palantir. I'd never heard of it, but a quick search on t'inter tells me that it's a proprietary bit of software, and I can't see anything in their website blurb that I've not seen elsewhere - significantly, it says that it is being used by 'a' (i.e., one) top pharma company. I would put money on this meaning that their sales team have managed to get under the guard of one company and has given them a trial licence, possibly free of charge with a lot of analytical support, to try and get a toe-hold into an industry that already has plenty of tools for this purpose.

I'm not saying that you're wrong: I would be sincerely (and professionally) in your debt if you would bung down something that confirms what you say and corrects my misapprehensions.