Review Low-Dose Naltrexone: What is the Evidence? A Narrative Review, 2026, Gouda

[Sasha]

I don't know what Jonathan has in mind, but there is one unsubtle case that I know you are aware of @Sasha.

If you have good blinding and a good cohort size, and find that about 15% of the treatment group report a useful level of improvement, then you need to look at the rate of improvement on the placebo. If about 15% improved on the placebo too, then you can't say that the treatment benefitted a subset.

There's nothing subtle about those statistics, though!

 

[Jonathan Edwards]

There's nothing subtle about those statistics, though!

The placebo trial example is a simple one. Things get more complicated when you start analysing things like the causal beliefs of groups of scientists by doing statistics on the results they report. You can then show that things like variance of the data across studies cannot in a million moons really come out like that if the beliefs in causation are justified. You could probably do the same on patient beliefs in causation relating to their treatments. There are interesting patterns in the F&M ritux phase 2 open label extension that could be analysed, for instance. If you manipulate beliefs (maybe unintentionally) you can get effects that can be shown statistically but are always hard to prove for individuals.