An antiviral shortened recovery in one trial and did not cut deaths in another

Two of the largest trials run during the first year of the pandemic tested the same antiviral in the same kind of patient and were reported, in much of the coverage, as contradicting each other. They did not. They asked different questions, and the difference between the questions is most of the story [1][2].

The drug

Remdesivir is an antiviral developed by Gilead Sciences. It imitates one of the building blocks a virus uses to copy its genetic material, so that copying stalls when the imitation is used. It had shown activity against coronaviruses in laboratory work before 2020, which is why it was among the first candidates tested when the pandemic began. Emergency authorisation in the United States came in May 2020, followed by conditional approvals elsewhere [4].

The first trial and its measurement

ACTT-1 was a double blind randomised trial in adults in hospital with COVID-19 and signs of infection in the lower airways. Double blind means neither the patient nor the treating team knew who was receiving the drug and who was receiving an inactive drip. A total of 1,062 people were randomised, 541 to the antiviral and 521 to placebo [1].

Its main measurement was time to recovery, defined as leaving hospital or staying only for infection control reasons. Median time to recovery was 10 days on the antiviral against 15 days on placebo, with a P value below 0.001. Deaths by day 29 were 11.4 percent against 15.2 percent, a difference the trial was not sized to test on its own [1].

The second trial and its measurement

The World Health Organization's Solidarity trial asked a blunter question: does anyone die less often. It enrolled patients at hundreds of hospitals across dozens of countries and randomised them among whichever of four repurposed drugs were available locally, or to no trial drug. There were no placebos, so everyone knew what they were receiving [2].

The interim report, published in the New England Journal of Medicine in 2021, covered 11,330 adults. Among those randomised to the antiviral, 301 of 2,743 died against 303 of 2,708 in the matched control group, a rate ratio of 0.95 with a P value of 0.50. The report concluded that the four drugs it tested had little or no effect on deaths in hospital [2].

The final report, published in The Lancet in 2022, had 8,275 people randomised to the antiviral or its control. Overall, 14.5 percent of the antiviral group died against 15.6 percent of controls, a rate ratio of 0.91 with a P value of 0.12, which does not clear the usual threshold. Among those not already on a ventilator the figures were 11.9 percent against 13.5 percent, a rate ratio of 0.86 with a P value of 0.02, which does. Among those already ventilated there was no benefit [3].

Why the two readings sit together

Time to recovery and death are not the same measurement, and a drug can move one without moving the other enough to detect. A trial of 1,062 people can measure a five day difference in recovery and be far too small to settle a question about mortality. A trial of 8,275 can settle the mortality question and never record how quickly people got better [1][3].

Blinding differs too. In an open trial, knowing who received the drug can shape when a doctor discharges someone or moves them onto a ventilator. That is one reason the larger trial chose deaths as its measure: it is the outcome least open to being nudged [2][3].

The timing matters as well. The blinded trial ran first. By the time the larger one finished, standard care in hospital had changed, most importantly with the routine use of a cheap steroid in patients needing oxygen. A drug added to better background care has less room to show a difference [3].

What the record supports

The final Solidarity report also combined its own results with every other randomised trial of the same drug in hospital patients. Read that way, the honest summary is narrow: a modest reduction in deaths among patients who are in hospital but not yet ventilated, no benefit once a person is on a ventilator, and a faster recovery observed in the one large blinded trial that measured it [1][3].

That is a smaller claim than the early coverage made and a larger one than the later coverage allowed. Neither trial was wrong. They were built to answer different questions, and the answers only look contradictory if the questions are collapsed into one.

What this does not settle

None of these trials describes what the drug does against the versions of the virus circulating now, in a population with widespread immunity from vaccination and previous infection, or in people treated outside hospital. Every figure above comes from patients sick enough to be admitted, in 2020 and 2021, under the standard care of that time [1][2][3].

The general lesson outlives the drug. When two large trials of the same treatment appear to disagree, the first thing to check is not which one to believe. It is what each one measured, who was blinded, and what else was being given at the time [1][2][3].