The question arrives in the hospital pharmacy about once a week, usually from a ward, occasionally from my own family. Does magnesium work. As asked it has no answer, because magnesium is not one intervention. It is nine or ten different interventions that happen to share a divalent cation, given by different routes, in different salts, at doses separated by a factor of thirty, and the evidence behind them runs from excellent to entirely absent.
Two things before the table.
The first is a distinction I will keep repeating until it is boring. Correcting a measured deficiency and supplementing a person who is already replete are two different experiments. Different mechanism, different expected effect size, usually a different literature. Almost every confused argument about magnesium is two people discussing these two things while using one word.
The second: I do not give individual advice, here or anywhere I cannot see the notes and the creatinine. What follows is about trials.
D and none are separate columns. The difference between them is where most of the shouting happens.
| Indication | Best evidence | Design and size | Effect | Grade | Comment |
|---|---|---|---|---|---|
| Eclampsia prevention in pre-eclampsia | Magpie Trial, Lancet 2002 | RCT, 10,141 women, 175 hospitals, 33 countries | eclampsia 0.8% vs 1.9%, a 58% lower risk (95% CI 40 to 71) | A | Intravenous magnesium sulphate, prescribed as a drug. Not a capsule. |
| Correcting documented hypomagnesaemia | physiology, plus Huang and Kuo, J Am Soc Nephrol 2007, for the potassium mechanism | no placebo trial, and none is owed | repletion fixes the deficit, and the hypokalaemia that will not correct without it | A for the indication | You are treating a measured number rather than a hope. |
| Migraine prophylaxis | Peikert 1996 and Pfaffenrath 1996, both in Cephalalgia; AAN and AHS guideline 2012 | two RCTs, 81 and 69 patients, 12 weeks each | attacks down 41.6% vs 15.8% in one; response 28.6% vs 29.4% in the other | B | Graded Level B, probably effective. The salts differed. See below. |
| Blood pressure | Zhang et al, Hypertension 2016 | meta-analysis, 34 double-blind RCTs, 2,028 participants, median 368 mg/day for 3 months | systolic 2.00 mmHg lower (95% CI 0.43 to 3.58), diastolic 1.78 lower (0.73 to 2.82) | B for the effect, D for it being enough on its own | Real, replicated, and roughly a fifth of what one antihypertensive does. |
| Glucose parameters in diabetes or at high risk | Veronese et al, Eur J Clin Nutr 2016 | meta-analysis, 18 double-blind RCTs | fasting glucose SMD 0.40 lower (95% CI 0.80 lower to 0.00), I² = 77% | C | The interval touches zero and the heterogeneity is enormous. |
| Idiopathic and nocturnal leg cramps, older adults | Garrison et al, Cochrane 2020 | 11 RCTs, 735 participants | cramps per week 9.59% lower (95% CI 23.14% lower to 3.97% higher); responders RR 1.04 (0.84 to 1.29) | D, moderate certainty | Asked and answered. |
| Rest cramps in pregnancy | the same review | 5 RCTs, 408 women | conflicting | none | Lower certainty, genuinely unresolved. |
| Insomnia, adults 55 and over | Mah and Pitre, BMC Complement Med Ther 2021 | meta-analysis, 3 RCTs, 151 participants | sleep onset latency 17.36 min shorter (95% CI 27.27 to 7.44); total sleep time 16.06 min longer, not significant | C | 151 people, every trial at moderate to high risk of bias. |
| Sleep in adults with no sleep complaint | none | none | unknown | none | This is the population the advertising addresses. |
| Constipation | magnesium oxide and hydroxide as osmotic laxatives, in use over a century | established | works | A | And this is the mechanism that caps your dose for everything else in this table. |
Two trials, one year, one journal, opposite results.
Peikert and colleagues randomised 81 patients with a mean of 3.6 attacks a month to 600 mg of magnesium as trimagnesium dicitrate daily or to placebo for 12 weeks, after a four week prospective baseline. In weeks 9 to 12 attack frequency fell 41.6% against 15.8% on placebo. Diarrhoea in 18.6%, gastric irritation in 4.7%.
Pfaffenrath and colleagues randomised 69 patients to 10 mmol of magnesium as the aspartate hydrochloride twice daily, also 12 weeks. Response rates came out 28.6% against 29.4% and the trial was stopped early.
You can read that as a contradiction. I read it as evidence that the word magnesium is not a specification. The salt differs, the counter-ion differs, the dissolution differs, and so does the dose: Peikert's 600 mg is given as 24 mmol, which is about 583 mg of elemental magnesium, while Pfaffenrath's 20 mmol is 486 mg. The 2012 guideline from the American Academy of Neurology and the American Headache Society graded oral magnesium Level B, probably effective, and I think that grade has held up better than most Level B grades from 2012.
That dose is not free. Nearly one participant in five had diarrhoea, because 583 mg sits well above the 350 mg upper level set for supplemental magnesium.1
The Cochrane review is the one place in this whole subject where I can hand someone a finished answer. Eleven randomised trials, 735 participants, five in pregnancy and five in idiopathic cramp. For older adults with nocturnal cramp the confidence interval on the primary outcome runs from 23.14% fewer cramps per week to 3.97% more, and the authors conclude that meaningful prophylaxis is unlikely at any dose used, at moderate certainty.
Moderate certainty of no useful effect is a finding. It is not the absence of one. Across those trials 11% to 37% had gastrointestinal side effects, so the intervention is not free either.
Pregnancy is the exception and the review says so plainly: conflicting, lower certainty, unclear. If someone tells you the Cochrane review settled cramps in pregnancy, they read the title.
Everything sold to you about magnesium and sleep rests on three randomised trials totalling 151 adults aged 55 and over who already had insomnia, pooled by Mah and Pitre in 2021, at doses from 320 to 729 mg of elemental magnesium daily for 20 days to 8 weeks. Sleep onset latency came out 17 minutes shorter than placebo; total sleep time 16 minutes longer and not significant. Every included trial carried moderate to high risk of bias and the certainty was low to very low. The authors say outright that the literature is not adequate for physicians to make well informed recommendations.
Notice the population. Older adults with diagnosed insomnia. There is no adequately powered trial of magnesium for sleep in healthy adults who sleep normally and would prefer to sleep better, which is precisely the person the packaging is talking to. So the cell is empty rather than negative, and an empty cell should be said out loud instead of being filled in with a mechanism story about NMDA receptors.
Nobody prints this table, so here it is: molar mass from standard atomic weights, elemental fraction by mass, and what 500 mg of the compound itself delivers.
| Compound | Formula | Molar mass (g/mol) | Magnesium by mass | Elemental Mg in 500 mg of compound |
|---|---|---|---|---|
| Magnesium oxide | MgO | 40.30 | 60.3% | 302 mg |
| Magnesium chloride hexahydrate | MgCl₂·6H₂O | 203.30 | 12.0% | 60 mg |
| Trimagnesium dicitrate, anhydrous | Mg₃(C₆H₅O₇)₂ | 451.11 | 16.2% | 81 mg |
| Trimagnesium dicitrate nonahydrate | Mg₃(C₆H₅O₇)₂·9H₂O | 613.25 | 11.9% | 59 mg |
| Magnesium bisglycinate | Mg(C₂H₄NO₂)₂ | 172.42 | 14.1% | 70 mg |
| Magnesium malate | MgC₄H₄O₅ | 156.38 | 15.5% | 78 mg |
| Magnesium L-threonate | Mg(C₄H₇O₅)₂ | 294.50 | 8.3% | 41 mg |
| Magnesium sulphate heptahydrate | MgSO₄·7H₂O | 246.47 | 9.9% | 49 mg |
A bottle marked "500 mg magnesium citrate" can mean 500 mg of the salt, which is 59 to 81 mg of magnesium depending on hydration state, or 500 mg of elemental magnesium supplied as citrate, which is six to eight times as much. Both labels exist and both are legal. A figure in millimoles would settle it in one character, and almost nobody prints one.
Oxide has the highest magnesium fraction on that list and the worst behaviour in the body. Walker and colleagues gave 46 healthy adults 300 mg of elemental magnesium daily for 60 days as citrate, as an amino acid chelate, or as oxide: 24 hour urinary excretion was higher on the two organic forms than on oxide (p = 0.033), and oxide was indistinguishable from placebo. Kappeler and colleagues gave 20 healthy men single doses of citrate and oxide in a randomised crossover and found serum magnesium higher on citrate at 2, 3, 4, 5 and 6 hours, with urinary excretion higher too.
But the salt is not the biggest lever. The dose per sitting is. Fine and colleagues fed eight healthy men a standard meal with 0, 10, 20, 40 or 80 mEq of magnesium acetate and watched fractional absorption fall from 65% at the lowest intake to 11% at the highest. Splitting the same daily amount across meals will do more for you than switching counter-ion, and unlike switching counter-ion it is free.
Serum magnesium is a weak instrument and it is the only one your laboratory offers. Roughly 0.3% of body magnesium is in serum; most is in bone and inside cells. The familiar 0.75 to 0.95 mmol/L interval came from the distribution in a normal population in NHANES I, not from clinical outcomes, and Costello and colleagues argued in 2016 for a lower bound of 0.85 mmol/L instead. A low result means something. A normal result does not exclude depletion. The magnesium loading test is the reference standard and essentially nobody runs it outside research. A test only works inside the assumptions that make it a test, which @ledger/benfords-law-and-when-it-lies-to-you demonstrates better than I can for a different measurement.
Magnesium is cleared by the kidney, so renal function decides whether the same dose is a laxative or an arrhythmia. Wakai and colleagues reported in 2019 on 320 patients prescribed magnesium oxide at a Japanese university hospital: 23% reached a serum magnesium of 2.5 mg/dL or above, and the risk factors on multivariate analysis included an eGFR at or below 55.4 mL/min (odds ratio 3.1) and a dose at or above 1,650 mg/day (1.9). Japan's regulator issued a safety notice on hypermagnesaemia with long term magnesium oxide in 2008 after reports that included deaths, and repeated the request for serum monitoring in 2020. Available without prescription is a description of a shelf, not of a pharmacology.
The deficiency case is strong and dull. Migraine is decent, specific, and limited by your bowel rather than by the evidence. Cramps are finished. Sleep is an advertising budget with three small trials behind it, run in a population that is not the one being advertised to. And the best magnesium evidence in all of medicine is an infusion given by obstetricians to women with pre-eclampsia, which tells you nothing whatever about a capsule.
The question is never whether magnesium works. It is whether you are short of it, and for what.
The 350 mg figure is not a cliff edge. The Institute of Medicine derived it from a lowest observed adverse effect level of 360 mg/day of non-food magnesium, with osmotic diarrhoea as the critical effect and an uncertainty factor of about 1, on the reasoning that diarrhoea is obvious to the person having it and reverses. It is also the only upper level set for supplemental sources alone rather than total intake. Costello, Rosanoff, Nielsen and West argued in Advances in Nutrition in 2023 that it is too conservative and ignores differences between salts. ↩