A cross-sectional survey was administered from May 6 to 17, 2018, to adult (≥18 years of age) Blued app users who were male at birth and located in Beijing, Tianjin, Sichuan, or Yunnan, China. Recruitment was conducted through the built-in advertising functions within the Blued app, which included pop-up messages, clickable links, banner advertising, and text message notifications. Text message notifications were sent to 631,963 randomly selected Blued users. Advertisements were designed by the study team and implemented by collaborators at Blued. A total of 34,701 Blued users clicked the link to the survey webpage. Advertisements asked potential participants if they would like to participate in a short research survey. Participants who clicked on the advertisements were taken to a survey platform designed by Sojump within the Blued app, which provided information about the study and a way to navigate to the eligibility screener. All eligible participants were consented electronically before completing the online survey, which was administered using the Sojump survey platform. Although 6040 Blued users initiated the survey process, 1368 of them did not complete the informed consent or did not meet the screening criteria, leaving 4672 participants eligible for these analyses. For analysis, data were exported from Sojump and stored on an encrypted computer at Blued in Beijing.

Ethical approval was provided by China’s National Center for AIDS/STD Control and Prevention (NCAIDS) (KX180117492), which is registered with the US Office for Human Research Protections (IRB0000227) and has a Federal Wide Assurance (FWA00002958).

In the survey, participants were asked to describe their last sex acts involving anal sex with condom use and without condom use, respectively. This allowed us to create a “case interval” (ie, last act of unprotected anal sex) and a “control interval” (ie, last act of protected anal sex) for each individual participant. The survey also collected other characteristics of the sex act (ie, interval-level characteristics), including where participants initiated the partnership (online [eg, through Blued or another app or website] or offline [eg, through friends, at a bar, in a park]), the partnership type (ie, main or committed partner, casual partner with multiple sex acts, or casual partner with a single sex act [one-time]), the participant’s knowledge of the partner’s HIV status (ie, negative, positive, or unknown status), the participant’s role in anal sex (ie, receptive, insertive, or both), and the participant’s engagement in substance use (eg, alcohol or illicit drugs) before sexual intercourse.

The survey also collected data regarding the participant’s demographic characteristics, sexual identity, substance use in the past 6 months (rush poppers, methamphetamine, methylenedioxymethamphetamine, gamma-hydroxybutyrate, ketamine, and ecstasy), sexually transmitted infections (STIs) in the past 6 months, and HIV status.

A total of 4672 participants completed the survey. Of those, 3361 participants (71.9%) were excluded from this analysis because they reported not having anal sex in the last 6 months (n=1215), never having condomless anal sex (n=267), always having condomless anal sex (n=1873), or currently using PrEP (n=6). This left a remaining sample of 1311 participants (28.1%) who reported data required for the case-crossover analysis: engaging in both protected and unprotected anal sex. Table 1 presents data on demographic characteristics, sexual orientation, drug use, HIV status, and STI status by inclusion in the analysis. The median age of the participants was 27 years (IQR 23-33 years). More than one-half (2751/4672, 58.9%) of the participants had completed college or more, 20.4% (952/4672) were students, and 71.9% (3360/4672) were employed. Most (3330/4672, 71.3%) of the participants identified as being homosexual. Compared with participants excluded from the analysis, participants included in the analysis were less likely to have attended college, were less likely to be students, were more likely to identify as homosexual, were more likely to report engaging in substance use in the past 6 months, were more likely to have had syphilis in the past 6 months, were more likely to have had gonorrhea in the past 6 months, and were more likely to be HIV positive.

Table 2 compares the distribution of interval-level characteristics in partnerships initiated offline and those initiated online within unprotected anal sex acts and within protected anal sex acts. Of the 1311 most recent unprotected anal sex acts, 1019 (77.7%) were initiated online. Of the 1311 most recent protected anal sex acts, 1097 (83.7%) were initiated online.

We found similar bivariate associations across case (unprotected sex) and control (protected sex) intervals. In both intervals, partnerships initiated offline were less likely to be with one-time partners (unprotected anal sex acts: 24.7% vs 52.6%, P<0.001; protected anal sex acts: 30.8% vs 54.9%, P<.001) and more likely to be with a partner who the participant believed to be HIV negative (unprotected anal sex acts: 42.1% vs 26.7%, P<.001; protected anal sex acts: 41.6% vs 23.2%, P<.001) than partnerships initiated online. In partnerships initiated offline, the participant was less likely to only engage in receptive anal sex (40.7% vs 49.3%, P=.02) and more likely to engage in substance use prior to having sex (10.7% vs 6.7%, P=.04) than in partnerships initiated online in the protected anal sex interval.

Table 3 presents data on associations between interval-level covariates and unprotected anal sex. In multivariable analyses, the belief that partners were HIV negative (adjusted odds ratio [aOR] 1.57, 95% CI 1.09 to 2.27, P=.02) was associated with unprotected anal sex, compared with not being sure about the partner’s HIV status. Our primary outcome, initiating a partnership offline, was positively associated with unprotected anal sex (aOR 2.66, 95% CI 1.84 to 3.85, P<.001), compared with initiating a partnership online.

We conducted four sensitivity analyses in order to assess other potential biases. First, not including partnership type as a covariate in multivariable analyses had virtually no effect on the results (Multimedia Appendix 1). Second, the associations between initiating a partnership offline and unprotected anal sex stratified by partnership type were all positive and were not significantly different from each other (Multimedia Appendix 2). Third, restricting the sample to participants who met sexual partners both online and offline had virtually no effect on the results (Multimedia Appendix 3). Fourth, restricting the sample to participants whose self-reported HIV status was either negative or unknown had virtually no effect on the results (Multimedia Appendix 4).

In numerous previous assessments observing app users and comparing them with nonapp users, including several studies conducted in China, the causal model proposed was that exposure to a dating app caused an increase in HIV risk behaviors, which in turn caused increased HIV incidence [11-15,17,22]. We hypothesize that an underlying factor might make individuals who download, install, create an account, and use a dating app inherently different from those who do not: intention to have unprotected sex. In this concept, the app is a venue (akin to a bar or other meeting place), with intention to have unprotected sex causing both app use and subsequent unprotected sex with a higher HIV transmission risk. This analysis provides the first data to take into account the intention to have unprotected sex by those using dating apps by using a case-crossover design. Our primary finding—consistent across bivariable, multivariable, and sensitivity analyses—is that individuals were more likely to have unprotected anal sex in partnerships that they initiated offline compared with partnerships that they initiated online. The relationship was substantial, with over 2.5 times increased likelihood of engaging in unprotected anal sex in partnerships initiated offline compared with those initiated online.

Although our findings are in contrast to previous assessments that were mentioned, there are also studies that have reported similar results. For example, a study among young MSM using Grindr (a geosocial online dating app) found significantly higher rates of condom use with partners met on Grindr relative to partners that they met elsewhere [18]. However, these same authors reported in a subsequent study that familiarity with Grindr (using it for at least 1 year and meeting more partners through Grindr in the past month) was associated with increased sexual risk behavior (ie, condomless anal intercourse with their most recent Grindr-met partner) [16]. Specific to China, a study found that MSM who used partner-seeking mobile apps (eg, Jack’d, Grindr, Blued) did not report more condomless sex than men who did not use apps [19]. The current study improves upon these previous studies by looking within app users and employing a case-crossover design to avoid making potentially biased comparisons.

We also found that participants in a large majority of partnerships (68.0%) were not sure about their partner’s HIV status. There are two potential explanations for this. One explanation is that there is a high proportion of Chinese MSM who do not know their HIV status, which was reported in two previous studies, with 39.1% [25] and 33.2% [26] of MSM having never been tested for HIV. In this study, nearly one-third (30.3%) of the 4672 MSM from the entire sample had never been tested for HIV, and more than one-half (57.1%) had not been tested for HIV in the previous 6 months. Another potential explanation for the high frequency of sexual events among partners with unknown HIV status is that there is limited communication about and disclosure of HIV status among MSM prior to having sex, which has been reported previously [27]. Furthermore, partnerships were more likely to feature unprotected anal sex when a participant believed that his partner was HIV negative (compared with participants reporting that they were unsure of their partner’s HIV status). This type of safer sex negotiation in light of perceived risk has been reported previously in numerous studies [28,29].

We acknowledge several limitations in our study. First, data were cross-sectional and required recall from the previous 6 months (although we targeted a specific event in that time span, potentially reducing recall bias). Recall bias, if any, would be expected to bias toward the null. Second, data were based on self-report, and unprotected anal sex may have been under-reported. Again, this would be expected to bias toward the null. Third, although the comparisons made were within Blued app users and interval-level covariates were controlled for in multivariable analyses, there is the possibility that even though app use appears to be associated with condom use, this may not generalize to all sexual risk behaviors. For example, app users could have an elevated risk of HIV transmission through other means, such as by having more sexual partners.

These limitations notwithstanding, ours is the first study to take into account the inherent intentions of individuals using geosocial networking apps, compared with those not using such apps, when looking at their effect on sexual risk behavior. Our assertion is that the factor motivating individuals to download, install, create an account, and use a dating app (ie, intent to have unprotected sex) contributes to the associations between app use and sexual risk behavior reported in previous studies. These findings have important policy implications. If dating apps are believed to contribute to increased risk of HIV transmission, then these apps might be perceived as a problem to be addressed by health agencies. However, if instead apps are a proxy for, and not a cause of, increased HIV risk, which our results suggest, then dating app platforms should be viewed as a useful venue to identify individuals at increased risk for HIV transmission to allow for targeted service provision such as HIV testing, condoms, and PrEP.