We used data from local jurisdictions participating in YRBS to assess changes in sexual behaviors and HIV/STI testing and from county-level gonorrhea and chlamydia case report data from CDC's AtlasPlus to assess changes in STI diagnosis rates. The YRBS is a biennial survey of health-related behaviors and experiences conducted in-person in school among representative samples of 9th to 12th grade students in the United States by state, territorial, or local departments of health and education.^16,17 AtlasPlus is an interactive tool that includes US surveillance data on chlamydia and gonorrhea case reports by county, diagnosis year, age, race/ethnicity, and sex.¹⁸

We used data from the COVID-19 School Data Hub (CSDH; https://www.covidschooldatahub.com/) to determine the predominant school learning model (in-person vs virtual) for the 2020–2021 school year in 28 local jurisdictions that participated in the 2019 and 2021 cycles of YRBS. These data report learning models (virtual, hybrid, or in-person) and, in some cases, the number of students participating in each model for school districts or individual schools. We prioritized district-level data¹⁹ when available (13/28 districts), followed by school-level data²⁰ for all schools including 9th to 12th grade students in the relevant school district(s) (15/28 districts). When CSDH data were only available for part of the school year or were otherwise insufficient to make a determination, we used legal epidemiological methods similar to previous CDC COVID-19 policy analyses^21,22: reviewing governmental websites for relevant school, executive, or administrative policies and confirming with additional sourced documentation from district websites and local news. When a single YRBS jurisdiction included multiple school districts or we relied on school-level data and the districts or schools within the jurisdiction reported different learning models, respectively, we attempted to identify a predominant learning model across districts/schools for the YRBS jurisdiction.

“Virtual” jurisdictions typically reported virtual learning from the beginning of the 2020–2021 school year (or shortly thereafter) through at least February 2021 with some moving to hybrid or in-person learning toward the end of the school year. “In-person” jurisdictions primarily reported in-person learning for the entire school year. Two jurisdictions were excluded from analyses because the multiple school districts in each YRBS consortium reported conflicting models for the majority of the school year and therefore no predominant model could be determined. Supplemental Digital Content 1, http://links.lww.com/OLQ/B202 provides detailed information regarding these determinations.

Self-reported sexual behavior outcomes included the following with denominators in parentheses: ever had sexual intercourse (all respondents), sexual intercourse in the past 3 months (ever had sexual intercourse), and >1 sex partner in the past 3 months, condom use at last sexual intercourse, and alcohol or drug use before last sexual intercourse (sexual intercourse in the past 3 months). Self-reported testing outcomes included ever having been tested for HIV and testing for an STI other than HIV in the past 12 months (ever had sexual intercourse). Ever having sexual intercourse was determined by the question, “Have you ever had sexual intercourse?” or, if that question was not asked, by reporting ≥1 lifetime partner when asked, “During your life, with how many people have you had sexual intercourse?”

For each YRBS jurisdiction included in our analysis, we extracted rates per 100,000 of reported chlamydia and gonorrhea diagnoses among 15- to 19-year-olds, stratified by sex and race/ethnicity, from the corresponding county in AtlasPlus. When YRBS jurisdictions were a single county, this was a 1:1 match. For city-level YRBS jurisdictions other than New York City, we used data from the county where the city is located; in all such cases, the YRBS jurisdiction was the county seat and comprised ~25% to 85% of the county's population (Supplemental Digital Content 1, http://links.lww.com/OLQ/B202). To generate citywide rates for New York City, we used the average of the rates among 15- to 19-year-olds for each of the city's five counties weighted by the proportion of the city's 15- to 19-year-olds residing in each county based on 2020 US Census data.²³

Local jurisdictions participating in YRBS in 2019 and 2021 were included in the analysis if it was possible to determine the predominant learning model for the 2020–2021 school year and data were available in the public-use dataset or upon request from the jurisdiction.

To assess the effect of virtual learning on our outcomes, we used difference-in-differences (DID) analyses to estimate the relative changes in each outcome from 2019 (preexposure) to 2021 (post-exposure) between jurisdictions where high school students primarily attended school virtually (exposed) and jurisdictions where they primarily attended school in-person (control) during the 2020–2021 school year. In all regression models, the DID term was parameterized as the interaction between year (2019 vs 2021) and learning model (virtual versus in-person).

Sexual behavior and testing outcomes were treated as binary (either yes vs no or yes vs no/don't know or not sure). We used Rao-Scott χ² tests to assess overall changes from 2019 to 2021 in all included jurisdictions and logistic regression for DID analyses. We estimated the percentage of respondents reporting each outcome by survey year and learning model, the absolute change in these percentages from 2019 to 2021 within each learning model, and the absolute difference in these 2 percentages (virtual minus in-person change; an estimate of the DID), both overall and stratified by self-reported sex and race/ethnicity. Due to limited power to detect interactions with smaller sample sizes, we conducted regression models only among sub-groups representing ≥10% of the total study population. Where jurisdictions did not ask the question(s) informing an outcome or its denominator in either cycle, we excluded this jurisdiction from analyses for that outcome.

We conducted two sets of sensitivity analyses. The first addresses the YRBS survey administration period changing from spring 2019 to fall 2021 in 14 (58%) of the 24 included jurisdictions, more frequently among in-person (5/8 = 63%) than virtual jurisdictions (9/16 = 56%).¹⁷ These shifts resulted in changes in calendar months being considered during recall periods for certain outcomes, overall age distribution, and within-grade age distribution. To generate DID estimates without the potential bias introduced by these differential changes, we conducted sensitivity analyses restricted to the 10 jurisdictions that administered YRBS during the same semester in both cycles. Second, we explored whether sex of sexual contacts impacted sex-stratified estimates. Given the relatively small sample sizes for respondents with same-sex contacts and our interest in STIs, we defined categories as female respondents with female contacts only, female respondents with male contacts, male respondents with female contacts only, and male respondents with male contacts.

To test the parallel trends assumption underlying DID analyses (outcomes in virtual and in-person jurisdictions would have followed similar patterns in the intervention's absence²⁴), we visually compared overall trends in outcomes from 2013 to 2021 (except STI testing, for which data were unavailable before 2019) between virtual and in-person learning in a subset of 10 jurisdictions (6 virtual, 4 in-person) with available YRBS data, comprising 53% of respondents included in DID analyses (Supplemental Table 5, http://links.lww.com/OLQ/B203, Supplemental Digital Content 3, http://links.lww.com/OLQ/B204). The YRBS analyses account for complex sample design.²⁵

Sexually transmitted infection diagnosis rates were analyzed at the jurisdiction level using Poisson regression models with robust variance, both overall and stratified by sex. We also calculated the difference between within-learning-model percent change in rates from 2019 to 2021 and the difference in these percent changes (virtual minus in-person learning). Due to the potential impact of changes in STI testing on diagnosis rates, we conducted a sensitivity analysis for overall rates adjusting for the proportion of 15-year-olds to ≥ 18-year-olds who reported past-year STI testing among those who reported ever having sexual intercourse in the corresponding YRBS jurisdiction and survey year. We tested the parallel trends assumption using overall diagnosis rates from 2013 to 2021 because county-level rates for 15- to 19-year-olds were unavailable before 2019 (Supplemental Digital Content 4, http://links.lww.com/OLQ/B205).

Analyses were conducted in SAS® OnDemand for Academics. P values less than 0.050 were considered statistically significant. The University of Washington Human Subjects Division determined that these secondary analyses of deidentified or aggregated data did not meet the definition of human subjects research.

We included 24 of the 28 local jurisdictions participating in YRBS in 2019 and 2021 for a total of 47,661 and 40,647 respondents in each survey year, respectively. The predominant learning model for the 2020–2021 school year was virtual for 16 jurisdictions, comprising approximately two-thirds of respondents in the study sample in both survey years (Table 1). All jurisdictions with predominantly in-person learning were located in the South, whereas virtual learning jurisdictions were distributed across all four regions with large proportions of respondents from the Northeast and West. Similar proportions were non-Hispanic Black across both years and learning models (26.0–28.0%). In both years, respondents from virtual learning jurisdictions were likelier to be Hispanic/Latino and Asian and those from in-person learning jurisdictions likelier to be non-Hispanic White. Distributions of sexual orientation were similar between learning models, but respondents from both learning models were less likely to be heterosexual in 2021 (71.3–72.3% vs 80.4–80.9% in 2019).

Across all jurisdictions, sexual behaviors remained stable from 2019 to 2021 except for a significant decrease in ever having sex and in both HIV and STI testing (Supplemental Table 1, http://links.lww.com/OLQ/B202 in Supplemental Digital Content 2, http://links.lww.com/OLQ/B203).

Overall, condom use at last sex among those who had sexual intercourse in the past 3 months increased from 53.2% in 2019 to 56.4% in 2021 in virtual learning jurisdictions but decreased in in-person learning jurisdictions from 57.0% to 51.5%, resulting in a significant relative increase in condom use associated with virtual compared with in-person learning (DID odds ratio [OR], 1.42; 95% confidence interval [CI], 1.12–1.79; Figure 1A; Table 2). In stratified analyses, this association was significant only among female and Black respondents, although similar patterns were observed among male respondents (Table 3).

Overall, STI testing among those who ever had sexual intercourse decreased in both learning models, but more so in virtual learning (29.2% to 18.0%) than in-person learning (20.0% to 15.1%) jurisdictions, resulting in a significant relative decrease in STI testing associated with virtual learning (DID OR = 0.75, 95% CI = 0.59–0.96) [Fig. 1B; Table 2]. In stratified analyses, this association was significant only among female and White respondents, though similar patterns were observed for Black respondents. Overall, reports of ever having HIV tested among those who ever had sexual intercourse followed a similar pattern (virtual learning decreased from 29.9% to 20.3%, in-person from 23.4% to 17.9%), but the relative change was not significant except among White respondents.

DID estimates for ever had sexual intercourse, sexual intercourse in past 3 months, multiple sex partners in past 3 months, alcohol or drug use before last sexual intercourse, and ever tested for HIV were not significant in overall analyses (Table 2). Stratified analyses are presented in Supplemental Tables 2 (sex), 3 (race and ethnicity), 4 (same semester of survey administration), and 5 (sex and sex of sexual contacts) [all in Supplemental Digital Content 2, http://links.lww.com/OLQ/B203].

In sensitivity analyses limited to the 10 jurisdictions (7 virtual, 3 in-person) that administered YRBS during the same semester in both years, DIDs for condom use, HIV testing, and STI testing were similar to overall estimates. For the remaining outcomes, estimates indicated relative decreases associated with virtual learning in samesemester jurisdictions that were not or less apparent in overall analyses, although this relative decrease was only significant for alcohol/drug use before last sexual intercourse (DID: OR, 0.64; 95% CI, 0.45–0.93). In sensitivity analyses exploring the impact of sex of sexual contacts, results were similar between female respondents overall and female respondents reporting male sexual contacts and between male respondents overall and male respondents reporting female sexual contacts only. The parallel trends assumption appeared to hold for all outcomes except condom use at last sexual intercourse (Supplemental Table 6, http://links.lww.com/OLQ/B203; Supplemental Digital Content 3, http://links.lww.com/OLQ/B204).

Across all jurisdictions, the mean reported diagnosis rates per 100,000 15- to 19-year-olds decreased for chlamydial infection and increased for gonorrhea from 2019 to 2021 (Supplemental Table 1, http://links.lww.com/OLQ/B202). Mean chlamydia diagnosis rates decreased in virtual and in-person learning jurisdictions from 2019 to 2021, but by a greater proportion in virtual learning jurisdictions (3,209 to 2,393 per 100,000 15- to 19-year-olds, a 25% relative decline) than in-person jurisdictions (2,359 to 2,054 per 100,000; 13% decline) (Fig. 1C; Table 4). Mean reported gonorrhea diagnosis rates remained stable in virtual learning jurisdictions (744 vs 740 per 100,000 in 2019 and 2021) but increased by 20% in in-person learning jurisdictions from 519 to 622 per 100,000 (Fig. 1D). These changes resulted in significant relative declines in both chlamydia and gonorrhea diagnosis rates associated with virtual learning (DID: incidence rate ratio [IRR] for chlamydia, 0.86; 95% CI, 0.84–0.88 and gonorrhea: IRR, 0.83; 95% CI, 0.79–0.87; both P < 0.0001). Adjusting for past-year STI testing did not affect these estimates [adjusted IRR for chlamydia = 0.86 (95%CI = 0.84–0.88) and gonorrhea = 0.82 (95%CI = 0.79–0.87); both P < 0.0001]. Findings were similar stratified by sex, though reported diagnosis rates were higher among female 15- to 19-year-olds than their male counterparts across years and learning models. The parallel trends assumption appeared to hold for both diagnoses (Supplemental Digital Content 4, http://links.lww.com/OLQ/B205).