2 ms·
1. It's a success of his IRB. Of the many requirements he described, both by his IRB and the auditors, he didn't make the effort to try to understand how any of
by euyyn 9y ago
1. It's a success of his IRB. Of the many requirements he described, both by his IRB and the auditors, he didn't make the effort to try to understand how any of them could be good for his subjects. Instead he focused on how they affected his study, and tried to find the bare minimum he had to do to get approved. If the result of it is he doesn't want to deal with IRBs anymore, it might be for the best.
2. I believe this study could have been done with much less personal effort. His hypothesis is the screening question leads to unwarranted diagnosis. He already had Dr. W. convinced that it's plausible, and willing to do full diagnosis to his patients. And he had other doctors that would only use the screening question. All that's required to refute the null hypothesis is compare the rate of "yes" vs "no" in the set of Dr. W's diagnoses, with the rate of the doctors that just use the screening question. If his hypothesis is true, the other doctors would get more positives. That's 1 bit of information he had to extract anonymously from the patients' existing records, which is a strict subset of the methodology he developed for the study.
- skybrian 9y agoIf you don't test both kinds of diagnostics on the same patients, that doesn't sound the same at all? As another example, if you have a ten-question survey, asking 1000 people one question is not the same as asking 100 people all ten questions. You don't learn nearly as much about the correlation between answers to different questions.
- euyyn 9y agoAssume the patients are assigned to the doctors in a way that's not affected by whether they have the condition or not (e.g. at random). If the screening question doesn't lead to over-diagnosis, then you'd expect the same ratio of positives/negatives from both sets. You need more subjects to attain the same statistical power, but the data is much easier to obtain. 100 subjects already sounded like overkill anyway.
- YawningAngel 9y agoIt doesn't strictly pertain to his hypothesis, but presumably you would like data on false positives and negatives as well as overall outcome percentages.
- euyyn 9y agoGranted. But to get that you need to jump many more hurdles (which are there for good reasons).