Zolpidem

The honest efficacy picture comes from the data submitted for approval rather than from later marketing. Pooled across 13 studies and 4,378 participants, the drugs shortened time to fall asleep by about 22 minutes against placebo, and showed no significant effect on wake after sleep onset, number of awakenings, total sleep time, sleep efficiency or subjective sleep quality ^[9]. The authors' own summary is that the drug effect and the placebo response were each small, and that together they produced a reasonably large clinical response.

That last point is worth carrying. In a 24-week trial, 89.8 percent on the drug said it helped them sleep against 51.4 percent on placebo ^[10]. The drug beat placebo clearly; more than half of placebo patients also reported it helped.

⚠️ THE COMA CLAIM IS THREE ERRORS AT ONCE. Two placebo-controlled series in disorders of consciousness found response rates of about 5 percent, 4 of 84 in the larger one, with no clinical feature predicting who would respond ^[14][13]. Responses typically lasted one to two hours and sometimes ended in increased sleepiness. The response appears bimodal: non-responders showed no partial effect at all. So the population is disorders of consciousness rather than coma, the frequency is about one in twenty, and the duration is an hour or two. It is a real and strange phenomenon, and every popular version of it overstates all three.

⚠️ THE BOXED WARNING IS ABOUT COMPLEX SLEEP BEHAVIOURS, and it earned its place. Reviewing 26 years of reports, regulators found 66 cases of sleepwalking, sleep-driving and similar states resulting in serious injury or death: 20 deaths including carbon monoxide poisoning, drowning, falls, hypothermia and motor vehicle collisions, and 46 serious injuries including third-degree burns and gunshot wounds ^[7]. ⚠️ The number that changes behaviour is that 22 of those 66 had already had a previous episode on a Z-drug, which is why a prior episode is now a contraindication rather than a warning.

⚠️ Sixty-six spontaneous reports over 26 years cannot give an incidence rate, and should not be read as one. The warning rests on severity and on that prior-episode signal.

Next-morning impairment depends on time, not sex. Taken in the middle of the night and tested about five hours later, 10 mg measurably impaired real on-the-road driving ^[8]. At eight hours, driving studies show no impairment in either sex ^[5]. Both are true; the interval is the variable.

Falls are a real and quantified hazard. Among hospital inpatients, those given zolpidem fell at 3.04 percent against 0.71 percent of those prescribed it but not given it, with an adjusted odds ratio of 4.37 after extensive adjustment ^[12].

Dependence is documented despite the drug not being a benzodiazepine. A national pharmacodependence network found abuse and dependence sufficient to change the country's own drug monograph ^[11].

The benzodiazepines of the 1960s and 1970s worked and brought dependence, next-day sedation and a poor reputation with them. The Z-drugs were the answer: chemically unrelated compounds acting at the same receptor site but preferring the subtype that carries sedation, on the theory that this would separate sleep from the rest. Zolpidem, an imidazopyridine, arrived in 1992 and became the most prescribed hypnotic in the world. Its later history is a series of narrowings. The subtype selectivity turned out to be modest against alpha-2 and alpha-3. The efficacy turned out to be mostly about falling asleep rather than staying asleep. Dependence appeared anyway. And in 2019 a boxed warning arrived for a rare class of sleep behaviours that had been accumulating in reports for a quarter of a century.