By Jon Kamp

While Covid-19 deaths head lower, raising hopes that the U.S. is turning a corner as vaccinations continue, states around the country are steadily finding previously unreported deaths that are causing data confusion.

The issues largely involve systems that states are using to try to report Covid-19 data in near real time, and not deaths reported more slowly through death certificates. These front-line numbers are the ones that fuel state dashboards and data trackers, like the closely watched one created by Johns Hopkins University, which help policy makers and the public closely monitor pandemic trends.

Ohio in February announced more than 4,000 additional deaths while reconciling its data, and Indiana added about 1,500. Smaller revisions have also recently come from Virginia, Minnesota and Rhode Island. On Thursday, authorities in West Virginia said medical providers hadn’t properly reported 168 deaths to the state’s public-health department.

“Nobody likes surprises, and nobody likes data that’s wrong because that’s what drives decisions,” said Ayne Amjad, West Virginia’s state health officer.

These issues underscore ways in which Covid-19 can still challenge data-reporting systems in the U.S. Like many countries, the U.S. is trying to track pandemic events nearly as they happen, and a big part of this effort has required speeding up how deaths are reported.

In West Virginia, reporting deaths would typically require waiting many weeks for death certificates to be completed, Dr. Amjad said. But the state last year asked medical providers to also fill out a one-page report for Covid-19 deaths to create a faster record. The state discovered the recent undercount, of all deaths from December and January, by using death certificates to determine that the 168 death reports weren’t properly filled out, Dr. Amjad said.

She said the reporting issues were at about 70 sites, mostly hospitals and long-term-care facilities. A Covid-19 surge like the one that hit the U.S. this winter can slow reporting, she said, but she and Gov. Jim Justice called the reporting errors unacceptable. The U.S. has seen more than 530,000 Covid-19 deaths, about half of them since Thanksgiving, according to data compiled by Johns Hopkins University.

On Tuesday, Minnesota health officials said an audit found four private laboratories failed to report lab results, and that led to finding another 138 deaths. These had been recorded on death certificates, a health-department spokesman said.

An audit in Indiana uncovered 1,507 historical deaths, mostly from 2020, state authorities said in early February. Death certificates were used to verify those fatalities, a spokeswoman for the state health department said. Soon after, an issue with unreconciled mortality data led Ohio’s health department to find 4,000 unreported Covid-19 deaths.

In Virginia, it was a system problem that recently led the state to add about 900 deaths. Officials there realized the number of deaths they were reporting didn’t seem to track a rise in cases, and death certificates helped correct the error, said Lilian Peake, Virginia’s state epidemiologist. “We realized something was wrong,” she said.

These state fixes aren’t filling in major gaps in what researchers believe is a significant undercount of Covid-19 deaths. This is underscored by a wide gap between known Covid-19 deaths and excess deaths, or deaths above average levels in recent years.

Misreporting Covid-19 deaths was particularly likely early in the pandemic, when testing was scarce and doctors filling out death certificates were less familiar with the disease, according to public-health experts. They have also chalked up some excess deaths to other issues, such as people avoiding hospitals during health emergencies.

“We’re kind of stuck with this underreporting, especially at the beginning of the pandemic,” said Robert Anderson, chief of the mortality-statistics branch at the National Center for Health Statistics, which is part of the Centers for Disease Control and Prevention.

Matching front-line surveillance data with death certificates can improve data in both systems, and can sometimes lead to rewritten certificates, Mr. Anderson said. But changing death certificates isn’t easy, he said. The person who filled out the death certificate—often a doctor—has to agree to amend the record.

“We are seeing some deaths that were not Covid before that had been attributed to Covid when amended, but it’s a relatively small number,” Mr. Anderson said.

The big changes at the state level can create at least temporary, and artificial, bulges in the data that Johns Hopkins and others are knitting together to show daily trends.

This briefly happened with the backlog of mortality data from Indiana and Ohio, before those were backdated, which Johns Hopkins tracks and reflects in its records when possible. There is still an artificially large, 469-death bulge in Iowa on Dec. 11, conversely, from when that state changed the way it reports Covid-19 deaths.

“This is the challenge, and this is why we need to work to improve our national surveillance,” said Jennifer Nuzzo, an epidemiologist and senior scholar at the Johns Hopkins Center for Health Security.