Sexual Risk and HIV Testing Disconnect in Men who Have Sex With Men (MSM) Recruited to an Online HIV Self-Testing Trial

AJ Rodger; D Dunn; L McCabe; P Weatherburn; FC Lampe; TC Witzel; F Burns; D Ward; R Pebody; R Trevelion; M Brady; PD Kirwan; J Khawam; VC Delpech; M Gabriel; Y Collaco-Moraes; AN Phillips; S McCormack


HIV Medicine. 2020;21(9):588-598. 

In This Article


SELPHI was an internet based, open-label, randomized controlled trial (2017–2019), which aimed to assess effectiveness of providing free HIVST kits to increase HIV diagnosis rates. The full trial methods have been published previously.[21] In brief, SELPHI had a two-stage randomization with a target of enrolling 10 000 participants. Randomization A took place at enrolment, with eligible participants randomly allocated (in a 3:2 ratio) to the offer of a free baseline HIV self-test (BT) versus no offer of a free baseline HIV self-test (nBT) (Figure 1). Randomization B was open only to participants who met further eligibility criteria[21] and were randomized to receive regular HIV testing reminders and the offer of a free HIV-self test kit (RT) versus no regular self-test (nRT).

Figure 1.

The HIV self-testing public health intervention trial schema.

The SELPHI trial uses the BioSURE™ (BioSure (UK) Ltd, Nazeing, UK) HIVST kit, an antibody immunoassay detecting HIV-1/2 antibodies from approximately 28 days after infection, requiring a whole blood sample from a finger prick. The HIVST kits were posted to participants by the manufacturer, using an address provided at enrolment.

Participants were recruited to the trial through sexual and social networking sites including Grindr, Hornet, Recon, Scruff and community Facebook webpages using advertising targeted to a broad spectrum of MSM and trans people which has been described previously.[22] Criteria for enrolment were: (1) age ≥ 16 years, (2) being resident in England or Wales, (3) being a man (including trans men) or trans woman, (4) ever having had anal intercourse (AI) with a man, (5) not being known to be HIV positive, and (6) having provided consent to link to the UK national HIV surveillance databases held by Public Health England. Very few trans women were recruited (n = 23) and they are not included in the current analysis and will be reported on separately.

This paper reports baseline data provided prior to randomization A. Data collected via an online survey at baseline included sociodemographics (gender, sexual identity, education, age, ethnicity and country of birth), recent sexual behaviour (numbers of AI partners and CAI partners in the previous 3 months), HIV testing history (time of last test, number of tests in the last 12 months and location of last test), sexually transmitted infection (STI) testing history, and PrEP and post-exposure prophylaxis (PEP) use. Postcode was collected to enable delivery of the HIVST kit. This enabled calculation of (geodetic) distance to the nearest genitourinary medicine (GUM) (level 3) clinic using geographical information system (GIS) mapping data. Travel time from a participant's postcode to the nearest GUM clinic (at peak hours on Monday morning) was estimated for public transport (TRACC programme; and for driving (ARCGIS PRO programme; The analysis utilized public transport time for London and driving time elsewhere.

The study protocol was approved by the UCL Research Ethics Committee (REC) and informed consent was sought from all participants (ref: 9233/001).

Statistical Analysis

Analyses were performed in STATA version 15.1 (StataCorp LLC, College Station, TX, USA). All analyses used data collected at baseline and prior to randomization. Two main analyses were performed. The first examined predictors of never having tested for HIV among all participants. The second examined predictors of not having tested in the 6 months prior to enrolment and included only men who reported two or more CAI partners in the 3 months prior to enrolment. The latter group are recommended to test quarterly according to UK guidelines, whether or not they are receiving PrEP. However, we used a conservative 6-month period for testing to allow for delays in appointments or attending.

Quantitative variables were classified into between three and five categories before associations with dependent variables were examined. Logistic regression analyses were based on participants with no missing data for any of the included variables (complete case analysis). Although this is less efficient than multiple imputation, the loss of efficiency is minimal as the frequency of missing data was low. For each variable, the category with the highest number of men was selected as the reference category. As ethnicity and being born outside of the UK were highly correlated, only ethnicity was included in the multivariable logistic regression models. Distance and travelling time to the nearest GUM clinic were also highly correlated, and the former was included in multivariable models. As the use of PrEP and the use of PEP were almost perfect predictors for having had an HIV test, these factors were not included in the multivariable model. The multivariable logistic regression models included all listed variables, regardless of the P-value from the univariable analysis, because parsimony is an irrelevant consideration when the number of observations greatly exceeds the number of variables.

Because of the very large sample size, many highly statistically significant associations were found, even when the size of the effect was modest. It is therefore more informative to focus on estimates and confidence intervals rather than P-values. Interpretation of the logistic regression models is focussed on the adjusted odds ratios, that is, after controlling for any confounding effects of the other factors in the models.