The T5K cohort study used established ([13]; also CG et al, unpublished data, 2019) internet-based strategies to enroll a large sample of HIV-negative men, transmen, and transwomen who have sex with men aged 16 to 49 years and are at high risk of HIV acquisition. The cohort will be followed prospectively for 48 months for the outcomes of PrEP uptake and HIV seroconversion. We aimed to enroll a cohort of participants at high risk for HIV that was geographically diverse (ie, representing every US state and territory), racially and ethnically diverse (4468/8777, 50.91% participants of color), and young (2221/8777, 25.30% aged 16-24 years). We achieved these goals without needing to employ stratified sampling.

Open enrollment for T5K began in October 2017 and concluded in June 2018, when 67,166 of the estimated 649,000 (67,166/649,000, 10.35%) males eligible for PrEP across the United States were using it [8]. Participants were recruited via ads on men-for-men geosocial sexual networking mobile phone apps (Figure 1). Although not the targeted audience, transgender women and men were not excluded if they reported sex with men and otherwise met the eligibility criteria (Table 1). The study was promoted as an opportunity to receive at-home, self-administered HIV testing. Advertisements were geotargeted to individuals using apps inside the United States and the US territories.

Potential participants were directed to a secure enrollment survey in their device’s Web browser and presented with a screen describing study participation and eliciting informed consent. The informed consent described the incentive schedule: US $15 for completing a secondary survey (ie, 1 after the enrollment survey) if they were eligible and another US $15 for completing self-administered at-home HIV testing (ie, oral fluid sample returned to the study laboratory for testing). Additional incentives, described in the informed consent, are available to participants who complete prospective longitudinal follow-up assessments.

Interested individuals were screened for eligibility via a Web-based survey collecting data on sexual behavior, substance use, demographic characteristics, history of PrEP use, and history of postexposure prophylaxis (PEP) use. The survey was programmed into Qualtrics survey software and tested by study staff. Measures had been previously used by the members of the research team or were derived from the published research. Our Web-based survey was divided into thematic blocks based on question content and used adaptive questions based on survey responses from the participants. Examples include survey questions about known HIV status, PrEP use, and main sexual partners. Survey items were not randomly ordered by participant. The number of items answered by the participants and displayed per page of the Web-based survey varied by subject and participant responses because of the skip and/or display logic used within the survey. However, to improve ease of use on mobile devices and reduce survey fatigue, each page contained 1 to 2 questions. Participants could not click back to view or change a previous response because of the skip and/or display logic depending on previous responses. If a participant chose an incorrect response, they could contact the study staff to reset that response. Before activating the survey, members of the study staff tested the survey extensively for usability and technical function. The survey was tested on Windows, Mac, iOS, and Android devices and on Chrome, Firefox, Internet Explorer, Opera, and Safari Web browsers. Eligible and consenting individuals were asked to provide contact information for longitudinal follow-up. All participants were assigned a unique identifier at study enrollment and this unique identifier was used for all study databases and datasets. Participants’ contact information was stored in an encrypted database separated from their questionnaire answers and other study-related information. Only designated study staff were allowed access to study databases.

Enrolled individuals were sent a secondary Web-based survey that assessed the psychosocial characteristics. As the study participation involved receiving and returning an at-home HIV test kit via mail, as well as follow-up HIV test kits, consenting participants were required to provide name, mailing address, email address, and other contact information. We followed established and effective measures to minimize repeat participation and fraudulent manipulation of HIV testing procedures, including recording internet protocol addresses of participants and using cookies to block repeated attempts (CG et al, unpublished data, 2019). Our enrollment survey blocked multiple submissions, our databases flagged duplicate contact information, and all mailing addresses were validated with the US postal service. Multiple entries were identified by email addresses and/or phone numbers. In addition, the data manager manually checked for duplicate entries during baseline data collection. We also assessed time to completion of our Web-based surveys and checked for variability in response sets.

Upon completing this secondary survey, participants were mailed an OraSure HIV-1 specimen collection device to use at home. Participants were also provided access to a study video along with printed instructions on completing the HIV test, as well as our phone number in case they had questions. Procedures involved taking an oral swab and placing it in an oral fluid container and mailing the specimen using provided prepaid shipping materials to the Wadsworth Center Laboratory of the New York State Department of Health for antibody testing (Avioq HIV-1 Microelisa System) and archiving. Participants indicated the date of collection and any samples received by the lab after 21 days were not analyzed. In these instances, participants were contacted to retest. Median number of days between specimen collection and lab receipt was 4 days (interquartile range 3 to 6 days).

Participants who enrolled and completed a baseline questionnaire but did not return an HIV test kit (baseline serostatus unknown) will continue to be followed and asked at regular intervals to submit an oral fluid sample for HIV testing using the at-home sampling kit. For the purposes of prospectively estimating HIV incidence in the T5K cohort, these individuals will be excluded. However, we will conduct post hoc sensitivity analyses that make assumptions about having similar, lower, or higher HIV risk profiles than the T5K participants for whom baseline HIV status was determined.

We estimated pre-enrollment HIV incidence using baseline data on HIV serostatus and HIV testing history from T5K participants. Using these data, we reconstructed a cohort of individuals who could all be classified as HIV negative as of 12 months before cohort enrollment. For those HIV positive at enrollment, we estimated the time of seroconversion using self-reported data on the timing of the last HIV-negative test. Specifically, those self-reporting a negative HIV test within 6 months of T5K enrollment were classified as being HIV negative as of 6 months before enrollment, with seroconversion timing assumed to be distributed evenly during the 6 months leading up to study enrollment. Similarly, participants self-reporting a negative HIV test 7 to 12 months before T5K enrollment were classified as being HIV negative as of 12 months before enrollment, with seroconversion timing assumed to be distributed evenly during the 12 months leading up to study enrollment. We also estimated HIV incidence for the 6-month period leading up to cohort enrollment. For the purposes of identifying boundaries around minimum and maximum incidence rates, we conducted sensitivity analyses representing assumptions of the timing of seroconversion at the extremes.

Enrollment in T5K had a targeted sample size of 5000 participants with a confirmed HIV-negative test for prospective follow-up. The 5000 number was chosen to provide precise estimates of HIV incidence of 1.11 per 100 person-years (193 seroconversions of 17,329 person-years of follow-up over 48 months). This sample size also allows 80% power to detect binary exposures with frequencies between 20% and 80% for adjusted relative hazards ranging from 1.73 to 2.27 as statistically significant.

All data from the Web-based surveys and the laboratory testing were imported, cleaned, and merged using SAS. Data were geocoded to an exact address or Zone Improvement Plan (ZIP) code. Maps were created in ArcGIS and did not include exact participant location.

The T5K study protocol was approved by the Institutional Review Board of the City University of New York (CUNY) Graduate School of Public Health and Health Policy.

In total, 43,161 individuals began our enrollment survey and 22,091 (22,091/43,161, 51.18%) completed it (Figure 2). Of the noncompleters (N=21,070), 61.04% (12,862/21,070) closed their browser window on the informed consent page (ie, immediately). Of the completers (N=22,091), 9193 (9193/22,091, 41.61%) were eligible; however, 1023 were excluded because we determined their response to be a duplicate entry. Of the remaining 8807 participants who provided informed consent to participate in the study, 30 were excluded, as they did not provide contact information.

The final sample was 8777 consented participants from all 50 US states, Puerto Rico, and Guam (Figure 3). The descriptive statistics of the cohort and HIV incidence rates calculated included responses from enrolled participants meeting the eligibility criteria and completing questionnaires.

The final cohort of 8777 individuals was geographically diverse, including participants from all 50 US states, Puerto Rico, and Guam (Figure 3). Nearly all (8554/8777, 97.46%) were cisgender male, 0.72% (63/8777) were transgender women, and 0.60% (53/8777) were transgender men who have sex with men (Table 2). There were also 107 (107/8777, 1.22%) individuals who self-identified outside of the gender binary—all reported being assigned male sex at birth. In total, 50.91% (4468/8777) were persons of color, and 25.30% (2221/8777) were young individuals aged 16 to 24 years.

Prospective closed follow-up of T5K participants includes completion of an annual self-administered at-home HIV testing and extensive Web-based surveys beginning 12 months after the baseline survey. In addition, participants will be contacted every 6 months (in between annual surveys) for a brief survey on HIV testing, diagnosis, and PrEP use (ie, attempts to access PrEP, PrEP initiation, PrEP adherence, and PrEP discontinuation). Participants who self-report being on PrEP are asked to provide proof in the form of a picture of the medication bottle with their prescription information. A comprehensive list of key study measurements by study wave is included in Table 5. We will follow the cohort of 8777 individuals for up to 4 years to characterize the following: the rate of PrEP uptake/discontinuation; individual /network /contextual-level determinants of PrEP uptake /discontinuation; patterns of PrEP use (eg, daily or on demand); the rate of new HIV seroconversions and other missed HIV prevention opportunities (ie, STIs while not on PrEP); individual/network/contextual-level determinants of HIV seroconversion and missed HIV prevention opportunities; racial/ethnic disparities in HIV incidence and their trends over time; and the influence of PrEP uptake on racial/ethnic disparities in HIV incidence.

Given the novel nature of this entirely Web-based nationwide cohort, we elected to longitudinally follow all participants providing consent at enrollment (N=8777), rather than only those who completed HIV testing. Participants who did not complete the secondary survey or subsequent HIV testing will have the opportunity to provide those data/samples at future assessments, and this will enable us to learn more about differential rates of participation and attrition in the cohort moving forward.

We have successfully recruited an entirely Web-based national cohort of confirmed HIV-negative men, transgender men, and transgender women who have sex with men and are at very high risk for HIV into longitudinal follow-up. Importantly, none of the T5K cohort members were on PrEP at enrollment but all met the objective criteria for PrEP use. This will allow our study to fill critical knowledge gaps that can improve the understanding of barriers to PrEP uptake and engagement among those most in need of HIV prevention interventions. This cohort also offers opportunities to examine the effect of different implementation strategies aimed at improving the uptake of and engagement with HIV prevention interventions, including those that can be delivered on the Web.

We observed a higher rate of several recent and lifetime STIs among persons with a history of PrEP compared with those who were never on PrEP. This could represent differences in the risk profile of persons who have used PrEP (ie, they were at higher risk for an STI in the past before or while using PrEP). However, many STIs are asymptomatic, making it likely that the STIs reported by this group at baseline would have been diagnosed as a result of them having initiated PrEP in the past, as PrEP services involve baseline and follow-up STI screening and treatment. We also observed that those with a history of PrEP were >8 times more likely to have taken PEP in the past. This could reflect better access to HIV prevention services or greater HIV risk among those with a history of PrEP use. Persons on PEP are usually good candidates for PrEP, and PrEP should be systematically discussed and offered to all persons completing PEP if there is a likelihood of ongoing HIV risk.

Our estimates of HIV incidence in the 12-month period leading up to T5K cohort enrollment (2.4% per year overall and 2.6% per year among those never on PrEP) suggest that the risk of HIV acquisition that we will observe prospectively will also be quite high, unless PrEP uptake increases dramatically. Indeed, we observed that HIV incidence was substantially higher among those never on PrEP compared with those with a history of PrEP before enrollment (IRR 2.36; 95% CI 1.21-5.20).

Major strengths of this cohort study include the large sample size, the wealth of self-reported information related to HIV risk and PrEP, direct measurement of HIV status and seroconversion, geographic representativeness, recruitment of participants independent of their access to/engagement with the health care system, and inclusion of large numbers of racial/ethnic minority GBM as well as those aged <25 years. Having every US state, Puerto Rico, and Guam represented allows for a robust exploration of state-level policies and other higher-level effects (ie, contextual factors) as potential determinants of PrEP uptake and HIV risk. More than 50% (4468/8777, 50.91%) of the cohort comprises HIV-negative men of color, allowing in-depth investigations into the mechanisms of racial/ethnic disparities in PrEP uptake, HIV incidence, and circumstances surrounding HIV seroconversion. Importantly, the T5K cohort study design, with semiannual Web-based at-home surveys and at-home self-sampling for HIV testing, reduces the potential for participation and questionnaire-response bias introduced by the Hawthorne effect, which could be stronger in face-to-face studies. Studies involving frequent face-to-face contact can cause participants to adopt behaviors that make them less representative of the high-risk populations from which they were drawn. Of note, McCabridge et al [16] introduced the construct of research participant effects that was built upon the Hawthorne effect [17] by elaborating on the implications of research on the mechanisms that introduce bias, including demand characteristics [18,19]. Studies involving high levels of staff contact with participants may induce behavior change by repeatedly engaging participants outside of their natural context, artificially influencing results [17,19,20] and reducing generalizability [21].

Recruiting participants and gathering data on the Web is a much less controlled research environment than face-to-face or telephone interview studies. Web-based recruitment also increases our vulnerability to repeat participation and fraudulent manipulation of HIV testing procedures (eg, someone else’s saliva, other than that of the enrolled participant, could be submitted to the lab). However, we followed established and effective measures to minimize these risks [22,23]. Although we took steps to assess whether participants were reading the interview questions, participants recruited on the Web may be less cognitively engaged in questionnaire completion and less inclined to provide a specimen than if they were interacting directly with study personnel. Our study population was recruited via sexual networking apps and represented a sample of those at high risk for HIV, who are not on PrEP at cohort enrollment. However, the underlying population that gave rise to the T5K cohort is not representative of all HIV-negative GBM at high risk for HIV in the United States [24]. Finally, HIV seroconversions and PrEP uptake may be incompletely ascertained, and the timing of HIV seroconversions must be estimated using a midpoint approach with broad intervals, potentially introducing bias in our baseline HIV incidence estimates. For example, our study eligibility/inclusion criterion of never diagnosed with HIV (self-report) effectively excludes frequent HIV testers who seroconverted and were diagnosed in the period immediately before study launch.

There have been many challenges and lessons learned in the launch and execution of the T5K study, which have been detailed elsewhere ([13]; also CG et al, unpublished data, 2019). In brief, although there are many advantages to entirely Web-based studies, challenges we have encountered include greater difficulty obtaining signed informed consent, returning HIV test results to participants, linking persons to HIV services when needed (especially in underserved areas), the potential for participants to be distracted, and difficulty ensuring unique and valid participants—and new challenges with regard to privacy and data security.

Intervention studies are of great interest to the T5K study team. Our goal is to utilize key findings from this cohort to develop interventions, potentially including some delivered completely on the Web, once the observational phase is completed. There will likely be opportunity for natural experiment design studies that allow for rigorous examination of future state or national policy changes, an introduction of long-acting PrEP formulations or dosing recommendations, and other novel HIV prevention modalities. To facilitate such research, we will gather extant data via geolinking with T5K participant information [25]. These include the participant’s ZIP code matched to county- and state-level data regarding, for example, STI rates, HIV incidence and prevalence, HIV viral suppression rates, residing in an Affordable Care Act Medicaid expansion state (yes/no), urban versus nonurban city/county location, state-wide PrEP policies, and pro-/anti-lesbian, gay, bisexual, and transgender policies, Human Rights Campaign state equality index [26], and the Movement Advancement Project (MAP) index [27].

T5K welcomes new collaborations. Instructions and a concept proposal form are available on our website or can be obtained by emailing the Principal Investigator (CG). Submitted concept proposals will be reviewed by CG and a core group of T5K investigators, with rapid turnaround.