Remdesivir, the first drug to be convincingly shown to help patients hospitalized with COVID-19, appears to work best before patients get so sick they need invasive breathing support.

The finding comes from a closely watched clinical study testing remdesivir against placebo, results of which were published Friday in the New England Journal of Medicine. The National Institutes of Health, which is running the trial, said last month treatment with the antiviral therapy hastened recovery from coronavirus disease. That led the Food and Drug Administration last month to issue an emergency authorization of the drug in the U.S.

Detailed data from the study confirm the NIH’s initial disclosure, showing hospitalized patients treated with remdesivir recovered four days faster than those given placebo. The drug’s benefit, however, was most pronounced in patients who required oxygen support, but did not yet need mechanical ventilation or a procedure known as ECMO that pumps blood through an artificial lung.

“Our findings highlight the need to identify COVID-19 cases and start antiviral treatment before the pulmonary disease progresses to require mechanical ventilation,” researchers wrote.

Physicians have generally expected remdesivir would be most beneficial when given earlier in the course of a patient’s disease, before immune system complications lead to impaired organ function and widespread inflammation. An antiviral, remdesivir is thought to work by blocking the virus from reproducing itself inside cells that it has invaded.

“Early on is when the virus is your problem,” said Katherine Perez, a pharmacy specialist in infectious disease at Houston Methodist hospital, in an interview following the NEJM publication of the results. “Once you’re ventilated, it’s your immune system.”

The full data from the NIH study appears to bear that hypothesis out, although the greater benefit observed in patients on oxygen but not ventilation could be a product, in part, of the larger sample size in the trial for that group. Study investigators also noted that the follow-up time reflected in these results may not have been long enough to capture the drug’s effect in the most severe patients.

“You may hit a point in which the degree of respiratory damage is so severe that the damage to the lungs is irreversible. This is not necessarily a remdesivir failure,” said Neera Ahuja, a division chief of hospital medicine at Stanford University School of Medicine, in an interview Saturday.

“On the other hand, it may be that remdesivir is just not effective at reversing damage to the lung after a certain point of time.”

The question of which patients would be helped most by remdesivir is an important one, particularly because supplies of the drug, developed and manufactured by Gilead, are tight. The biotech has said it will donate its currently supply of 1.5 million vials, which is enough to treat approximately 140,000 people using a 10-day treatment course.

While the NEJM results suggest ventilated patients aren’t as likely to benefit, the decision to give remdesivir or to withhold for another, less sick patient who might be helped more is still fraught.

“The cohort we’re really struggling with is the intubated patients,” said Perez. “With such a limited resource, can we justify using [remdesivir]? Can we justify not giving it?”

The NEJM paper gives physicians more information to make those decisions, but data from the NIH trial, and others, are still being collected.

The NIH made its study results public in late April after an independent monitoring committee reviewed the data and concluded the drug’s benefit was great enough to require disclosure. At that point, there was no statistically significant difference in the rate of death between the two trial groups.

The results published Friday still don’t show a clear survival benefit for remdesivir, although 22 fewer patients on the drug died than those given placebo. At two weeks, 7.1% of patients on remdesivir had died, compared to 11.9% on placebo.

Some have questioned the NIH’s decision to disclose results early and allow physicians to give remdesivir to patients who were initially randomized to receive placebo — potentially missing an opportunity to prove whether treatment helps more patients live.

Even if remdesivir does help, though, the drug on its own won’t be enough to keep in check a disease that’s led to the deaths of nearly 100,000 Americans and more than 338,000 people worldwide.

“Given high mortality despite the use of remdesivir, it is clear that treatment with an antiviral drug alone is not likely to be sufficient,” concluded researchers in NEJM.

More than 1,000 patients were enrolled in the trial, which was conducted at 60 hospitals and treatment centers across the U.S., Europe and Asia. Researchers moved with extraordinary urgency, taking less than three months to conduct the study and publish results.

The trial remains the only randomized, placebo-controlled study to support the use of a drug to treat COVID-19. Other studies have tested remdesivir, with one, conducted in China, showing no benefit to treatment. But that trial was much smaller than the one run by the NIH, and was stopped early after researchers there were unable to enroll enough patients.

The NIH will continue testing remdesivir, but it now considers the drug to be the standard of care against which other drugs should be compared. Study of a combination of remdesivir together with an immune-blocking drug called baricitinib is now underway.