I bring up the ethics question because a strict utilitarian analysis would be primarily answer the question of whether the drug is effective, looking for any re

I bring up the ethics question because a strict utilitarian analysis would be primarily answer the question of whether the drug is effective, looking for any red flags secondarily. A do no harm analysis (in the ethics sense, not the experimental design sense) has to look for any harms as a primary goal, maybe even equally to the question of efficacy. Answering the question of whether the experiment harmed anything important is always much harder, as I’m sure you know. You see some metric is way down, but after a multiple comparisons correction it might just as well be noise, so then you have to start digging, doing a bespoke analysis for the fact that these non iid metrics point in these particular directions. You can’t pre-automate it because with N metrics that are practically significantly up/flat/down, that’s at least 3^N possible investigations to convince yourself whether it’s enough to stop rollout. That’s the part that takes time, in any analysis I’ve ever done. Maybe literally zero time to check efficacy because that part is automated, then sometimes days to check whether there’s anything else worrisome in the probably-just-noise on other metrics, spending time proportional to the cost of getting the harm part wrong. Some of the hard questions aren’t statistical. You see some KPIs definitely up and some KPIs definitely down and deciding whether to roll out becomes a bitter fight between competing interests. Maybe you don’t roll it out yet for client segment X (like this trial and kids and pregnant women, as I understand it). Re US vs UK, no idea. I’m not saying 3 weeks is necessarily the right amount of time here. I’m just so taken aback at the idea that 3 weeks is necessarily too long to the point you “would probably have to start firing people” and would “absolutely” have to start putting them on PIPs.