Great commentary on the paper and RDD in general jamanetwork.com/journals/jam..., and podcast discussion edhub.ama-assn.org/jn-learning/...

All this suggests that routine Mg supplementation of everyone falling below the reference range may be another example of low value care. Very low levels and high-arrythmia risk patients likely still need treatment, but not everyone who's number is red on the EMR. RCTs needed.

1) Some CIs wide, and 2) Results only apply to subjects near the cutoff. However, vast majority of Mg supplementation is given to people who are only slightly below the cutoff, and result was consistent across the range of cutoffs we studied.

We did not find evidence that it did. There was no difference in the risk of tachyarrhythmias, shock, or death in the following 24 hours, comparing those just above and below the treatment cutoff. However, there's a few important limitations...

Patients either side of the cutoff differed significantly in their probability of Mg supplementation, but were otherwise well matched. How did this affect clinical outcomes?

We used multi-cutoff fuzzy regression discontinuity in a cohort of 171k ICU admissions to compare those just above and below the lower end of the reference range (depending on the hospital, anywhere from 1.6 to 2.0 mg/d, or 0.66-0.82 mmol/L)

journals.lww.com/jtrauma/full...

Diuretics may well ⬇️mortality in HF, we just don't know. But not for the first time, 'meta-analysis of RCTs' has laundered ambiguity and weak methods into apparently solid conclusions from the top of the evidence pyramid (4/4).

Now for the implausible mechanism. Of 12 deaths in the placebo arm, the commonest cause was cancer (4 deaths), vs. 1 cancer death in the diuretic arm. Anti-neoplastic effect of diuresis or random chance? You decide. (3/4)

First, the error. Cochrane mortality OR is based on 3/111 deaths w/diuretics vs. 12/110 w/placebo. But looking at the 3 source RCTs, there are in fact *4* deaths in the diuretic arm. Adding this, the OR remains just about 'significant', 0.30 (0.10-0.98), but very fragile (2/4)

It was all totally wasted. The findings were wrong and the net contribution to scientific understanding was negative. Publish or perish, ultimately pushed by funders and institutions, is driving a toxic culture of scientific waste. We have to stop. 5/5

Collectively the individual observational studies (plus the SRs, and the SR of the SRs) represents at least 10s of thousands of labour hours from researchers, ethics panels, funding reviews, journal reviewers etc., and a large amount of $$$. 4/

Once again we have to ask: what was the point of this observational research? There was an obvious risk of unmeasured confounding, and none of the studies had a way to address this. But they published anyway! 3/

Since 2020, at least 3(!) systematic reviews on this question (plus an SR of the SRs!). Pooled results were 0.65 (0.51-0.8), 0.71 (0.59–0.85), and 0.71 (0.63-0.79). There was an obvious risk of confounding by indiction: sicker patients -> more difficult IV -> IO . 2/