COVID-19 during pregnancy is associated with an increased risk of maternal death, intensive care unit admission, and preterm birth [1]; however, in the United States, uptake of COVID-19 vaccination during pregnancy is low [2]. Surveys indicated that the most common reason for refusing COVID-19 vaccination during pregnancy was a lack of safety data [3], which are limited because people who were pregnant were excluded from preauthorization clinical trials. The Centers for Disease Control and Prevention (CDC) recently released the first US data on COVID-19 vaccine safety in pregnancy, based on postvaccination health information reported by participants voluntarily enrolled in V-safe [4]. According to the CDC, although the preliminary data do not indicate any obvious safety signals, continued monitoring is needed, especially in early pregnancy and the periconception period (within 30 days before the last menstrual period). The CDC suggests that additional evidence of COVID-19 vaccine safety in pregnancy is critical for increasing the acceptance of COVID-19 vaccination in pregnant populations [2].

In the United States, 42% of people aged 18-29 years and 27% of people aged 30-49 years use Twitter [5]. Our prior work [6] demonstrated that Twitter data can be used to assess outcomes associated with medication exposure during pregnancy. Therefore, we hypothesized that Twitter could also be a source of data for assessing outcomes associated with COVID-19 vaccination received during pregnancy. Although user-generated Twitter data may be subject to potential limitations similar to those that the CDC has discussed regarding their participant-reported data [4] (eg, selection bias, reporting bias, misreporting, small sample size, limited information on other risk factors), the current availability of other sources of data is very limited. The objective of this preliminary study was to assess whether Twitter data could be used to identify a cohort for epidemiologic studies of COVID-19 vaccination in pregnancy. In particular, we explored whether it is possible to identify users who have reported (1) that they received COVID-19 vaccination during pregnancy or the periconception period, and (2) their pregnancy outcomes.

The Institutional Review Board of the University of Pennsylvania reviewed this study and deemed it exempt from human subjects research under Category 4 of Paragraph b of the US Code of Federal Regulations Title 45 Section 46.101 for publicly available data sources (45 CFR §46.101(b)(4)).

To facilitate a preliminary assessment of self-reports of COVID-19 vaccination on Twitter, we developed 6 handwritten, high-precision regular expressions designed to match tweets mentioning that the user received at least one dose of COVID-19 vaccination (Multimedia Appendix 1). In prior work [7], we developed an automated natural language processing (NLP) pipeline that detects tweets from the Twitter streaming application programming interface that announce a user’s pregnancy, and then collects all of their publicly available tweets on an ongoing basis. We deployed the 6 regular expressions on the collection of these users’ tweets that were posted through to the beginning of July 2021. To help determine if users were vaccinated during pregnancy (or the periconception period), we manually compared the timestamp of the tweets that matched the regular expressions with the timeframe of their prenatal period. To help estimate the timeframe of the users’ prenatal period, we drew upon an automated NLP tool, developed in our prior work [8], that uses a rule-based approach to search tweets for reports of the baby’s gestational age, due date, or date of birth, and extracts an estimate of the beginning and end dates of pregnancy based on the specific information in the tweet.

For users who posted tweets with a timestamp indicating they were vaccinated during pregnancy or the periconception period, we drew upon additional automated NLP tools, developed in our prior work [9-11], that use supervised classification to search tweets for reports of adverse pregnancy outcomes, including miscarriage, stillbirth, preterm birth, low birth weight, birth defects, and neonatal intensive care unit admission. To reduce the potential reporting bias in assuming that the lack of tweets self-reporting an adverse pregnancy outcome represents the lack of an adverse outcome, we also deployed an automated NLP tool, developed in our prior work [12], to search users’ tweets for reports that the baby was born at a gestational age of at least 37 weeks (ie, that the user is at least 37 weeks pregnant, or that the due date is in 3 weeks or less) and a weight of at least 5 pounds and 8 ounces. A gestational age of at least 37 weeks indicates the lack of miscarriage or preterm birth. A birth weight of at least 5 pounds and 8 ounces indicates the lack of low birth weight or, as a report of live birth, miscarriage, or stillbirth. If we did not automatically detect a tweet explicitly reporting a gestational age of at least 37 weeks, we manually analyzed tweets posted during this time for evidence that the user was still pregnant.

We manually verified the content of tweets detected automatically, identifying 150 users who reported on Twitter that they received at least one dose of COVID-19 vaccination during pregnancy or the periconception period. Table 1 presents examples of tweets that we used to identify these 150 users. For example, user 1 reported being 16 weeks pregnant on June 15, 2021, and therefore our automated tool [8] estimated that pregnancy began on February 23, 2021. User 1 reported receiving COVID-19 vaccination on March 24, 2021, which is approximately 1 month into the pregnancy. User 2 reported being 13 weeks pregnant on June 21, 2021, and our automated tool [8] estimated that the pregnancy began on March 22, 2021. User 2 reported receiving COVID-19 vaccination on March 6, 2021, which corresponds to the periconception period. The tweets in Table 1 also show that some users reported the vaccine manufacturer (eg, “#PfizerVaccine”) or dose number (eg, “second vaccine”), which can help distinguish mRNA vaccines from other types. Based on our estimates of the prenatal period for these 150 users, 90 (60.0%) of their pregnancies may have been ongoing. We manually verified at least one reported outcome for 45 of the 60 (75%) completed pregnancies. Table 2 presents the outcomes reported by these 45 users.