Two biases sit inside this design, and they pull in opposite directions.
The first is confounding by indication. A patient with twelve months of melatonin in the chart is, by definition, a patient whose insomnia did not resolve [7]. Both arms carried a chronic insomnia diagnosis [5], which removes the crudest version of that problem, and the investigators excluded anyone prescribed another sleep medication [6], which removes a second. What the release does not include is the list of variables the matching actually ran on [19].
The second bias runs the other way. The comparison arm was defined by the absence of any melatonin entry anywhere in the record [7], and in the United States melatonin sells over the counter with no regulation of strength or purity [9]. Some share of those nonusers are users whose bottles never reached a chart, and that pulls the two groups toward each other, making the observed gap an understatement of whatever real difference exists.
Then the arithmetic. Heart failure in 4.6% of users against 2.7% of comparison patients [3] is a gap of 1.9 percentage points [13]; as a ratio, it is 1.70, or 70% higher, not 90 [14]. That discrepancy hints that the headline figure comes from a time-to-event or covariate-adjusted model rather than the raw five-year proportions, though the release does not say which [19]. Run the absolute difference the other way and you get roughly 53 people on long-term melatonin for one additional heart failure case over five years [15]. The thing this doesn't tell you is anything about a few nights of melatonin for jet lag, which is the supplement's other common use [17].
The part I would keep is the second analysis. Requiring two prescription fills at least 90 days apart [8] trades sample size for confidence that the exposure actually happened, and the estimate settled at 82% rather than collapsing, which is how a signal behaves and not how chart noise usually does. The associations with heart failure hospitalization and death from any cause moved in the same direction [1], although those endpoints are not independent of the first.
Ekenedilichukwu Nnadi, the lead author and chief resident in internal medicine at SUNY Downstate/Kings County Primary Care, put the clinical implication in a conditional, saying that if the study is confirmed it could change how doctors counsel patients about sleep aids [10]. That is the right place for it, and none of this establishes cause [4]. The wider issue is that melatonin has been promoted as a safe sleep aid while the evidence on its cardiovascular safety over long stretches stayed thin [16], and heart failure already affects about 6.7 million US adults [12], so a modest shift in incidence among long-term users is the kind of thing that surfaces only in records mining this large. What would settle it is exposure measured rather than inferred: a dose and duration gradient inside a cohort where someone knows what was swallowed.
