What’s Wrong with CDC Data on Pediatric Mortality?

You have probably heard about the recent correction in pediatric death data by the CDC. This is related to issues I have been fighting about with the CDC since May 2021, and there has been a lot of confusion around the topic, so I wanted to provide an explainer.

Where are pediatric deaths tracked?

The primary issue stems from the fact that the CDC has multiple places where it tracks deaths by age, including the Data Tracker (updated 6 days/week) and various NCHS data files (updated weekly). In addition, the AAP tracks deaths by compiling data from state dashboards where available.

The CDC says the Data Tracker is more “real-time” but the NCHS numbers are more accurate. I believe the Data Tracker demographics page still has serious flaws, and should not be used. I recommend using the NCHS Provisional Deaths.

More information on data sources for pediatric mortality.

COVID Data Tracker – Demographic Trends

The CDC’s Covid Data Tracker is a large dashboard containing a variety of information to track the pandemic, including the new Covid Community Level metrics and various case, hospitalization, death, and vaccination data. The case and death data relies on surveillance data reported by the states. The top-level data for case totals is pretty consistent, but when you get down to more specific case/death data by demographics, the data completeness and accuracy falls apart.

Background on Data Tracker Demographic Issues

I first discovered the Data Tracker Demographics page when helping investigate with Emma Woodhouse on Twitter where some media sources were getting their numbers of pediatric deaths, because the numbers cited were larger than what were listed in the NCHS data. Here’s a thread about our investigation from May 2021. We inquired with the CDC, and were told the (lower) NCHS number should be considered the more accurate of the two, and that they were working to fix the discrepancies. Some changes were made, but the Data Tracker still was higher than the NCHS data.

There have been some additional revisions to CDC Data Tracker numbers over time, but they have remained consistently higher than NCHS numbers for children (and often young adults), while having significantly incomplete data compared to NCHS for adults 29 and older. The CDC always explains there is a short lag of a couple weeks to get data from the surveillance systems into NCHS, and sometimes longer for children, due to time it takes to investigate the death and submit the death certificate. However, they’ve never explained why adult deaths are severely UNDER counted in the Data Tracker compared to NCHS. It’s clear to me that the Data Tracker has incomplete data on adults because not all states provide complete data. But then why aren’t pediatric deaths just as incomplete?

The Data Tracker is missing almost 200K deaths compared to the NCHS database and the surveillance death totals shown at the top of the CDC Data Tracker page itself. This makes pediatric deaths appear to be a larger percentage of the overall total, and means anyone who wants to exaggerate the risk to children can use these larger unvetted numbers, under the guise of using the most “up-to-date” information. They dismiss the NCHS data as too lagged, even though there is no good evidence that the lag is nearly as significant as the difference between the Data Tracker and NCHS would make it appear.

Slides for CDC ACIP and FDA VRBPAC meetings use the more accurate NCHS numbers, but the Data Tracker numbers are often cited by experts on Twitter, in the media, by the AAP, and sometimes even by Dr. Walensky herself in briefings.

What’s behind the March 16th change to pediatric deaths?

Starting in late 2021, the difference between the pediatric deaths in the Data Tracker and NCHS began growing, and by late February, the Data Tracker numbers were really out of line, and continuing to grow at an unrealistic rate compared to NCHS numbers, and overall death reporting (which peaked nationally in early February). I brought this issue back up on Twitter and contacted the CDC again, because I was seeing more people on Twitter citing the growing pediatric death numbers from the Data Tracker to advocate for continued masking and other restrictions on children, even as the data began to make less and less sense. A week later, I issued a call to journalists to help me investigate and get some answers from the CDC, since they were not replying to me.

In March, Julia Raifman started saying that almost 1/3 of Covid deaths had occurred during 2022. She used this statistic in a Washington Post op-ed on March 8, and pitched the story to Melody Schreiber, who used it for a click-bait headline in The Guardian on March 11. Only problem is, it was based on faulty data, making the headline entirely untrue. Of course it spread like wildfire anyway. (The Guardian article has now been updated, but of course, the damage is done. The WaPo article has still not been updated at this time.)

After The Guardian article was published, the issue came to a head. I contacted the editors at The Guardian and a Twitter friend contacted the editors at Washington Post, who assigned their fact checkers to the issue. I also had a freelance writer that I had reached out to contact me for more details. I also had some other people I work with trying to reach out via FOIA requests and congressional pressure. On top of that, the Data Tracker pediatric death count increased from 1,567 to 1,755 in less than a week, and Melody Schreiber tweeted out that 188 more children had died since her article was written, 10% of all US child deaths during the pandemic. This was completely nonsensical – it’s more than an entire month of pediatric deaths from the peak of the Delta wave. With all of this going on all at once, someone at CDC finally realized they had a problem with the Data Tracker.

What did the CDC change about pediatric deaths on March 16th?

On March 16, the death counts on the Data Tracker Demographics page were updated, removing over 70 thousand deaths from across all age groups, but notably over 400 pediatric deaths (nearly 25%) of the previous total. They cited a “coding logic” error, in other words, a bad database query and/or programming code (not the cause of death codes on death certificates). A CDC spokeswoman later told the Washington Examiner that it removed “deaths previously reported across 26 states because CDC’s algorithm was accidentally counting deaths that were not COVID-19-related.”

This change only affects the numbers on the CDC Data Tracker Demographics page. It did NOT affect the total number of deaths reported by surveillance systems the US (the death total shown on the CDC web site, the NY Times tracker, JHU, Worldometers, etc.).

This is not based on an audit of how cause of death is determined, or anything about “from” vs. “with” Covid. It’s a single web page that had faulty numbers on it. The data did not match underlying data submitted by states to the CDC. It was a mistake that was fixed by a programmer, NOT an epidemiological determination about Covid-19 death definitions, comorbidities, or other cause of death decisions.

Was this intentional, or an accident?

While it’s impossible to say for sure, I believe this was an accidental error, but that the CDC should be held accountable for not identifying and fixing the issues sooner and more completely. The explanation from the CDC about a programming error is basically exactly what I suspected it was. Someone messed up the programming around importing, adding, and/or displaying data from the states. If the CDC were intentionally falsifying about this, they could have done a much better job. The recent increases were growing at an alarming rate, well after the Omicron wave had subsided, NCHS numbers are much smaller, and even data on the Data Tracker Demographics page directly below the deaths by age graph are much lower than what appears on that graph.

I do believe the CDC is guilty of terrible programming on this page, terrible quality assurance around the data displayed, and terrible follow up when people like myself asked legitimate questions about the accuracy of the data. If the data is used as “near real-time data in an emergency … to guide decision-making” the team at the CDC who actually uses this data either knew the data was junk and was ignoring the bad data, or should have been in an utter panic about pediatric deaths and raising all kinds of alarms. Neither option makes the agency seem competent.

I do think the CDC has been guilty of putting out a lot of MMWRs that serve as sheer propaganda to scare parents about the risks to kids, to support the value of masking in schools and in the community, and to push other narratives. But I think this the faulty Data Tracker demographics page is more about incompetence than malice (Hanlon’s razor). It’s reasonable to disagree on this opinion, but my point in bringing this issue to light is just to get the data fixed, something I still don’t think they have fully figured out.

Is the Data Tracker correct now?

No, I don’t think so. Pediatric deaths on the Data Tracker are still much higher than found in the NCHS data and the AAP report. Meanwhile, adult deaths on the Data Tracker are all significantly behind NCHS data. I still think there is some kind of an error that causes more pediatric deaths to be reported here – either counting some deaths that aren’t Covid at all or double-counting some deaths. In the week after the fix, the Data Tracker added 75% more pediatric deaths than NCHS. Due to NCHS lag, there are likely fewer than 12 children dying per week now, so Data Tracker deaths (which are supposed to be more “real-time”) should be increasing by *less* than NCHS, not more.

I will keep adding to this document as time allows, but wanted to get some information out there to answer questions I’ve been getting.