Unlike computer science where if you test something and it works once you and then you expect it to work for everyone, in medical research if you see an effect in one treatment study you probably won't see the same effect elsewhere.
For treatments (unlike controlled vaccine studies) the cohorts are too small and there are too many other variables.
People differ. Healthcare systems differ. Too many small details, too small to document, differ.
Some of the worst offenders have been scientists, who promote their own studies (eg, the French hydroxychloroquine studies).
It's possible to pick holes in these studies if you know the field. But these holes aren't pointed out in the papers themselves, and peer review doesn't stop this.
Peer reviewers are experts in the field. So they know if they are looking at a 30 person study it isn't conclusive, and wouldn't expect the author to state that. It's just assumed the readers know that.
We see the real impact of this with things like the Vitamin D studies. Experts know that Vitamin D deficiency is highly correlated with age and other health issues, but that the causation chain is "Person is unhealthy so doesn't get outside as much so gets less Vitamin D, and in some cases are deficient which can make health issues worse". Non-experts read the report and see the causation going the other way, ie, they think: "High Vitamin D fixes health issues".
Outside of the "hard" sciences, there is always a study that "proves" X and another that "disproves" it.
Not because scientists cheat, but because science is hard.