The Centers for Disease Control and Prevention estimated that the omicron variant now accounts for roughly 59% of all COVID cases in the United States, a significant decrease from the agency's previous estimate. The update shows how hard it is to track the fast-spreading variant in real time and how poorly the agency has communicated its uncertainty, experts said.

Last week, the CDC said that omicron accounted for approximately 73% of variants circulating in the United States in the week ending Dec. 18. But in its revision, the agency said the variant accounted for about 23% of cases that week.

In other words, delta, which has dominated U.S. infections since summer, still reigned in the United States that week. That could mean that a significant number of current COVID hospitalizations were driven by infections from delta, Dr. Scott Gottlieb, a former commissioner of the Food and Drug Administration, suggested on Twitter. Hospitalizations typically lag several weeks behind initial infections.

Experts said they were not surprised by the revisions, given that the CDC's estimates are rough guesses, with a wide range of possible values known as "confidence intervals." Cases of omicron can only be confirmed by genetic sequencing, which is performed on just a portion of samples across the country.

And omicron is still spreading extremely fast.

Still, they said the CDC did a poor job communicating the uncertainty of its estimates. The agency has suffered a series of black eyes during the pandemic, including sending out botched tests early on and shifting guidance on masking. On Monday, when it halved the recommended isolation period to five days for those who test positive but show no symptoms, critics objected that there was no requirement to test before returning to work.

David O'Connor, a virus expert at the University of Wisconsin-Madison, said, "The 73% got a lot more attention than the confidence intervals, and I think this is one example among many where scientists are trying to project an air of confidence about what's going to happen."

O'Connor said he initially thought the initial 73% point estimate "seemed high." The agency came up with the estimate based on a "relatively small number of sequences," he added.

"It's like playing Name That Tune and trying to say, based on just the first note, if the song is 'Ice Ice Baby' by Vanilla Ice or 'Under Pressure,'" O'Connor said. "Without more data, it can be really hard to know which one it's going to be."

The new estimate of 59% is also a rough calculation, experts said, and will most likely be revised in future weeks.

"I just want people to be very aware that that is an estimate; that's not actually from sequence-confirmed cases," said Nathan Grubaugh, a public health researcher at the Yale School of Public Health. "With omicron in particular, it's been very difficult to have any sort of projections, because things are changing just so so rapidly."

Grubaugh, who is tracking probable omicron samples in Connecticut, said that the variant makes up more than 80% of cases there, though he also notes that the country is heterogenous and the variant likely has a different prevalence in different places.

"I don't know how the CDC built their algorithm, but human beings made these programs, and humans are fallible," said Massimo Caputi, a molecular virus expert at the Florida Atlantic University School of Medicine. "At the end of the day, you can predict as much as you want, but you need to look at the numbers you have in your hand."

O'Connor, who is tracking omicron in Wisconsin, said the variant made up half the cases on the University of Wisconsin-Madison campus in just three days. "If I was making a betting prediction, it wasn't so much that the number 73% was wrong, but the timing to get there was wrong," he said.

These predictions will likely become more accurate over time as more data on omicron is collected.

More precise numbers will be needed to smartly distribute COVID treatments. One of the great challenges of omicron is the variant's ability to thwart two of the three monoclonal antibody treatments, which can prevent serious illness in COVID-19 patients. As such, some hospitals have begun scaling back these treatments; administrators at NewYork-Presbyterian, NYU Langone and Mount Sinai all said they would stop giving patients the two treatments that are ineffective against omicron. But the drugs could still help people infected with delta.

"If you still have those delta cases, discontinuing monoclonals means all those people who would have benefited from them won't be receiving them at all," O'Connor said.

O'Connor said scientists and health care providers need to do a better job of communicating the uncertainty in the predications they share with the public. "Having the humility to acknowledge that there's a lot that no one knows and is unknowable right now is going to be really important."

This article originally appeared in The New York Times.