OpenVSafe by OpenVAERS
Search V-safe

V-safe FAQ

What v-safe was, where this data came from, how we loaded it, and what it can and can't tell you.

What was v-safe?

v-safe was the CDC's smartphone check-in program for people who got a COVID-19 vaccine. It started in December 2020. After registering, people got text messages linking to short web surveys: daily for the first week after a dose, then weekly to six weeks, then at 3, 6 and 12 months. Parents and carers could register a child or another adult and answer for them.

Each survey had check boxes (fatigue, headache, fever and so on) and questions about whether the person could work or do normal activities, and whether they got medical care. There were also free-text boxes where people could describe what was happening in their own words. When someone said they had needed medical care, the v-safe call center could phone them and help file a report to VAERS, the federal Vaccine Adverse Event Reporting System.

Where did the data come from?

The CDC released the v-safe data in several parts, under the Freedom of Information Act. OpenVSafe loads all of them:

The data dictionary describes each part and its fields.

How did we load it?

We loaded the files as released into a database, then cleaned them into the tables the site reads. Cleaning means things like reading dates in one format, splitting combined answers into separate fields, and tagging each check-in with its survey (which dose and which day or month after it).

Every part of the release carries the same registrant code, a pseudonym the CDC gave each person (it looks like ABC-12345-67890). That code is what lets the site put one person's registration, doses, check-ins, free text, coding and calls on a single page.

Free-text search uses a full-text index. It finds the same entries as a plain "contains" search, but needs at least 4 characters.

How does the site count?

How are people matched to VAERS?

The release doesn't link v-safe people to their VAERS reports, so we matched them. For each call center case whose outcome says a VAERS report was filed, we took every v-safe VAERS report with the same vaccination date and manufacturer and a compatible sex and age, and scored each one. The score, out of 11, adds up timing (VAERS received the report soon after the call), dose and shared symptoms; state only breaks ties.

A match is high confidence when it scores 8 or more and at least 3 points ahead of the next-best report, and medium when it scores 5 or more and 2 ahead. A person's page shows a possible VAERS report only when the best match scores 8 or more.

What about deaths in the free text?

v-safe never asked whether someone had died, but families often used the free-text boxes to say so, usually on the person's own account ("Deceased 10/20/2021", "(b)(6) passed away on 3/23", "This is his wife, he died on May 6"). The Died filter on Search V-safe finds these people, and their page highlights the entry that reports the death.

To build it we took every free-text entry with a death word in it (died, death, deceased, passed away, funeral and so on): 5,648 entries. Then we sorted each one into four groups:

Word rules settled the clear cases. We read the 1,140 entries the rules couldn't settle, and checked the rest in samples, reading every entry the rules counted that had first-person or near-miss wording. Where someone wrote "my husband died" or "my mother passed away", the registrant's sex usually settles it. On a man's account, "my husband died" is almost always his wife reporting his death. On a woman's account, it is her own grief. A bare "I died" with no date or cause isn't counted, because many are jokes. "I died on April 6" or "I passed away" is counted. Someone "dying in hospice" isn't counted until the death is reported.

The result is 1,986 entries from 1,576 people. The filter also counts 186 more people with a MedDRA death code, checked as described below, for 1,762 people in all. It's a floor, not a count of deaths. Most people who died had no one to write it in, and some who did were never asked again. Like everything in v-safe, these are reports, not death records.

How were the topics on Search V-safe found?

Some things can't be found with one search term. People write "miscarried", "no heartbeat at my 8 week scan" and "lost the baby", and a search for "blind" also finds "blinding headache". So the Topics filters on Search V-safe were worked out ahead of time. For each topic we listed the ways people describe it and pulled every entry that matched. Word rules set aside the clear mistakes, and we read the 1,737 entries that could go either way. Every topic is about the registrant, not a relative or a friend.

We then added people with a matching MedDRA code. The CDC's coders worked from wording the search missed, and from text that isn't in the free-text release. Where a coded person's free text was available, it could overrule the code: we looked at each person the search hadn't found and left out those whose own words showed the code didn't fit, such as a husband's death coded as "death", or "blinding headache" coded as "blindness". People with no text to check count on the code alone.

These are floors, not rates. They show what people chose to write, and an entry doesn't show that the vaccine caused what it describes.

How do the pregnancy pages work?

The pregnancy pages combine the registry survey with everything else that points to a pregnancy: a pregnancy MedDRA code, a check-in answer of "pregnant: yes", or a call center pregnancy record. People who were 51 or older, or under 12, at their first dose are left out, unless they show a sign of pregnancy loss. Weeks of pregnancy at a loss or a premature birth are read from the free text ("miscarriage at 7 weeks", "born at 33 weeks"), since v-safe never asked.

What can't the data tell you?

What about privacy?

The release has no names, addresses or exact birth dates. People are identified only by their CDC registrant code, with an age band and the first three digits of their zip code. Some free text mentions personal details anyway. The CDC blacked out the parts it chose to, and we publish the entries as released.

For what the site collects about visitors, see our privacy policy.

OpenVSafe is a project of OpenVAERS, made possible by the support of Brownstone Institute and REACT19.

Loading…