Trends in HIV incidence between 2013–2019 and association of baseline factors with subsequent incident HIV among gay, bisexual, and other men who have sex with men attending sexual health clinics in England: A prospective cohort study – PLoS Blogs

Study design and participants

Methodological details of the study have been published previously [10]. The AURAH2 study was a prospective cohort study that recruited GBMSM who were HIV negative or of unknown HIV status from 3 large sexual health clinics in London and Brighton (56 Dean Street, London; Mortimer Market Centre, London; and Claude Nicol Clinic, Brighton) from July 2013 to April 2016. Participants were eligible if they were aged 18 years or older and had attended the study clinics for routine testing for sexually transmitted infections (STIs) or HIV. Men were classified as GBMSM for the purposes of the analysis if they met at least one of the following criteria: (i) reported being gay or bisexual; (ii) reported anal sex with a man in the past 3 months; or (iii) reported having disclosed to their family, friends, or workmates as being gay, bisexual, and/or attracted to men. Participants who consented to the study completed a confidential baseline paper questionnaire in the clinic. During the follow-up period, participants self-completed subsequent 4-monthly and annual questionnaires that were available online from March 2015 until March 2018. The baseline questionnaire gathered information on demographic, socioeconomic, lifestyle, health and well-being–related factors, knowledge and understanding of HIV, sexual behaviours, STI diagnoses, and PrEP and postexposure prophylaxis (PEP) use. The 4-monthly questionnaires assessed information on HIV status, HIV testing history, sexual behaviours, and lifestyle factors. Annual questionnaires captured the same information as the 4-monthly questionnaire and additional information on PrEP and PEP use in the past year, relationship status, and health and well-being factors as assessed on the baseline questionnaire. This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 Checklist).

Ethics approval and participant consent

All participants provided written, informed consent before taking part. Consent to participate in the study included consent for linkage to Public Health England (PHE)’s datasets at the end of the study using limited participant identifiers. The AURAH2 study was approved by the designated research ethics committee, The National Research Ethics Service (NRES) committee London-Hampstead, ref: 14/LO/1881 in November 2014 [10]. Based on the research protocol and all versions of study documents, the AURAH2 study subsequently received permission for clinical research at the 3 participating National Health Service (NHS) sites: Chelsea and Westminster NHS Foundation Trust, Central and North West London NHS Foundation Trust, and the Brighton and Sussex University Hospitals NHS Trust. The AURAH2 study was registered on the NIHR clinical research network portfolio.

Completion of online follow-up questionnaires

Participants who completed a first online follow-up questionnaire in March 2015 had the option to complete up to 9 online questionnaires, as the follow-up finished in March 2018. When participants were due to complete a questionnaire, 2 remainder emails were sent after 2 and 4 weeks followed by a text message. If participants missed a questionnaire at any time during follow-up, they were still invited to complete subsequent questionnaires. At each online follow-up, participants were asked about the most recent date of HIV test and the result.

Baseline measures

All baseline measures were self-reported in the participant baseline questionnaire. Sociodemographic variables included age group (<25; 25 to 29; 30 to 34; 35 to 39; 40 to 44; ≥45 years), country of birth and ethnicity (white UK born; other ethnicity UK born; white non-UK born; other ethnicity non-UK born), self-reported sexual identity (gay; bisexual/other plurisexual identities), education (university degree; other qualification; no qualification), ongoing relationship (yes, living with partner; yes, not living with partner; no), employment status (employed; not employed), sufficient money for basic needs (yes; mostly; sometimes or no), and housing status (homeowner; renting including private, housing association, and council; unstable or other).

We considered the following measures of sexual/HIV-related behaviour in the preceding 3 months (classified as “yes” or “no” unless otherwise indicated): condomless anal sex (CLS), number of CLS partners (none; 1; 2 to 4; 5 to 10; >10), CLS with partners known to be HIV positive, sexual CLS role (no CLS; always insertive; always receptive; insertive and receptive [versatile]), group sex, sex for drugs or money, fisting, or sex toys use. We also considered HIV test in the previous 6 months, and bacterial STI diagnosis, number of new sexual partners (0 to 10; 11 to 49; 50 to 99; ≥100), and PrEP and PEP use in the previous 12 months. Bacterial STIs included gonorrhoea, syphilis, and chlamydia, including lymphogranuloma venereum (LGV).

Lifestyle factors included recreational drug use (injection drug use; noninjection use of 1 or more of the 3 chemsex-associated drugs [mephedrone, GHB/GBL, crystal methampethamine]; non-injection use of other drugs; no drug use), smoking status (never smoked; ex-smoker; current smoker) and alcohol consumption (higher-risk alcohol consumption: a score of ≥6 on a modified version the AUDIT-C WHO alcohol screening tool questionnaire, first 2 questions only) [11]. A total score of 6 was chosen given that AURAH2 participants were only asked the first 2 questions of the WHO AUDIT-C questionnaire rather than the full AUDIT-C. Mental health included symptoms of depression (defined as a score of ≥10 on the Patient Health Questionnaire [PHQ-9], which is the standard cutoff score used to define clinically significant depressive symptoms) [12], and anxiety symptoms (defined as a score of ≥10 on the Generalised Anxiety Disorder Scale [GAD-7], which represents the standard cutoff to define anxiety disorder) [13].

For sexual/HIV-related behaviour, mental health, and alcohol consumption measures, missing responses were considered to indicate the absence of the event or condition, because our outcome of interests was “past report of behaviours.” Where there was no report of these measures, including missing, we classified answers as “no.” For all other variables that were not classified as “yes” or “no,” missing values were excluded from the analyses.

Time-updated measures

Age, recent HIV test, CLS, CLS with 2 or more partners, sexual CLS role, group sex, chemsex (a different variable from recreational drug use variable at baseline questionnaire; “have you used drugs before or during sex (chemsex) in the last three months?”, classified as “yes” or “no”), and bacterial STI diagnosis were also used as time-varying variables derived from baseline, 4-monthly, and annual questionnaires. Relationship status, PrEP use, PEP use, recreational drug use, injection drug use, depressive symptoms, anxiety symptoms, and alcohol use were time-varying variables derived from baseline and annual questionnaires. All other variables were fixed variables that were only ascertained at baseline.

Ascertainment of incident HIV

There were 2 methods of ascertainment of incident HIV diagnosis during follow-up. First, records of all GBMSM enrolled in the AURAH2 study were linked to national HIV surveillance data by PHE [14,15]. The databases collect information on new HIV diagnoses from laboratories, genitourinary medicine (GUM) clinics, general practitioners (GPs), and other services where HIV testing takes place in England. The data linkage process was carried out using a deterministic and hierarchical algorithm, based on gender identity, date of birth, year of birth, country of birth, ethnicity, originating clinic, years in the UK, and first initial and Soundex code (a 4-character coding of an adult surname). All data collected as part of the national HIV surveillance programme in the UK is pseudo-anonymised; no names are collected. The data matching process was completed in November 2019. For each study participant that matched to the HIV surveillance dataset, PHE data were provided on date and region of HIV diagnosis, CD4 and viral load at HIV diagnosis, and if relevant, time from diagnosis to linkage to care, time from diagnosis to treatment initiation, and death.

The second method of ascertainment of new HIV diagnoses was through the online follow-up questionnaires; participants were asked about the date and results of most recent HIV test. All the participants who reported being newly diagnosed with HIV in a follow-up questionnaire were also identified as having a new HIV diagnosis in the PHE surveillance databases. Linkage with PHE databases also identified a small number of participants who were positive at entry to the study (n = 3); these men were excluded from analysis.

Statistical analysis

For the analyses of HIV incidence and baseline-associated factors, all men enrolled in AURAH2 were included. Incident HIV infection was defined as seroconversion from HIV–negative status at baseline to HIV–positive during follow-up, confirmed by PHE. Person-years (PY) of follow-up were calculated from the date of completing the baseline questionnaire until (1) the date of HIV diagnosis from PHE for men who seroconverted or (2) 3 months before the date of data linkage with PHE datasets was completed (June 30, 2019) for men who did not seroconvert. Due to the linkage with PHE data for ascertainment of the endpoint, all men could be considered as remaining under follow-up over the entire period, even if follow-up questionnaires were not completed.

HIV incidence rates (IRs) were calculated as the number of new HIV infections divided by the number of PY of follow-up, reported with 95% confidence intervals (95% CIs). IRs were calculated per 100 PY, overall and according to calendar year from 2013 until 2019. As the study started on July 30, 2013 and ended on June 30, 2019, the first 2 years (2013 to 2014) and the last 2 years (2018 to 2019) were combined. The associations of baseline factors and current calendar year as a continuous variable with HIV incidence were analysed by calculating HIV IRs and using 2-level random-intercept proportional hazard models with sexual health clinic sites defining the second level to estimate conditional hazard ratios (HRs). The conditional distribution of the response given the random effects was assumed to be a Weibull distribution. HRs with 95% CI are presented unadjusted, and adjusted for sociodemographic factors that were less likely to be influenced by HIV and sexual behaviour: age at baseline, country of birth and ethnicity, sexual identity, and education.

Changes in the annual prevalence of sexual/HIV-related risk behaviours over time were also examined. The prevalence of CLS with 2 or more partners, group sex, bacterial STI, any recreational drug use, injection drug use, noninjection chemsex-related drug use (all in the previous 3 months), and PrEP and PEP use in the previous 12 months was calculated for each year from 2013/14 to 2018/19, using all available baseline and follow-up questionnaires from all participants at each time point. Trends over calendar time during the AURAH2 study period were assessed using univariate generalised estimation equation (GEE) models with a logit link and robust standard errors, accounting for multiple questionnaires responses from individual participants.

We also performed an additional longitudinal analysis among men who completed at least 1 online follow-up questionnaire to examine time-updated factors associated with HIV incidence. We used 2-level random-intercept Poisson regression models, unadjusted and adjusted for age (time-updated), country of birth and ethnicity, sexual identity, and education, using all available baseline and follow-up questionnaires. We present these results as incidence rate ratios (IRRs) with their corresponding 95% CI. In the multivariable analyses, the whole statistical unit for a single individual with missing values was excluded from the analyses if a value for one of the covariates was missing (complete case analysis).

All analyses were planned prior to analysing final datasets from PHE in November 2019 (S1 Analyses Plan), and no data-driven changes took place to these analyses, except that we used mixed-effects modelling instead of Cox proportional hazard modelling (indicated in the analyses plan), in response to peer review comments. The use of hierarchical models was chosen to take into account of clustering according to clinic. All analyses were conducted using Stata statistical software (version 15.1).

Characteristics of the participants

Between July 2013 and April 2016, a total of 1,162 HIV–negative men were enrolled in the study (Table 1). At baseline, the mean age of participants was 34 years (standard deviation [SD]: 10.4; interquartile range [IQR]: 26 to 39), 81.9% were of white ethnicity, 93.6% self-reported being gay, 74.4% had a university degree, 82.9% reported being employed, and 77.4% always had money to cover basic needs. In the previous 3 months, 63.9% reported having had CLS, 35.4% reported CLS with 2 or more partners, 12.7% reported CLS with HIV–positive partners, 43.1% reported group sex, 60.0% reported the use of at least 1 recreational drug, and 3.3% reported injection drug use. The type of drug injected was not ascertained in the baseline questionnaire, but all 38 people who injected drugs reported having taken at least 1 chemsex-related drug in the past 3 months. Overall, 38.0% of men reported having been diagnosed with a bacterial STI in the past year, and 5.0% and 20.7% reported ever having taken PrEP and PEP, respectively, in the past year. Three individuals did not complete the baseline questionnaire. The proportion of missing responses was low (<5% for all variables) (see footnotes in Table 1).

Of the 1,162 men enrolled, all were included in the PHE linkage for ascertainment of new HIV diagnosis. Of the 1,162 men, 622 completed at least 1 online follow-up questionnaire (54%), of whom 483 (78% of 622) completed at least 1 annual follow-up questionnaire, and 400 men (64% of 622) were followed until the end of the study. Men who were older, had greater financial security, with more stable housing, with university level education, and were employed were more likely to continue on the study (622 men versus 540 men who completed only the baseline questionnaire) (S1 Table). The number of follow-up questionnaires (4-monthly and annual) completed by the end of the study period was 3,277. Participants completed a median of 6 (IQR: 3 to 7) online questionnaires.

Trends in HIV incidence

In total, 33 of 1,162 men (2.8%) were newly diagnosed with HIV during the period from the date of completion of their baseline questionnaire until June 2019. Of all 33 diagnoses identified by the PHE linkage, 15 were self-reported by the participant on one of the AURAH2 online follow-up questionnaires. There were no additional unconfirmed self-reported HIV diagnoses. The 3 men who did not complete a baseline questionnaire were included in the incidence analysis as data on their age, HIV status, and PY of follow-up time were available from PHE. There were no deaths recorded among the 33 men diagnosed with HIV.

The overall HIV IR in this cohort with 4,618.9 PY of follow-up time was 0.71 (95% CI 0.51 to 1.00) per 100 PY (Fig 1 and Table 2). HIV incidence fell progressively from 2013 until 2019; from 1.47 (95% CI 0.48 to 4.57) per 100 PY in 2013/2014 to 0.25 (95% CI 0.08 to 0.78) per 100 PY in 2018/2019. The incidence declined on average by 0.85-fold per year from 2013 to 2019 (p < 0·001, modelled using mixed-effects Weibull proportional hazard).

Fig 1. HIV incidence among GBMSM in the AURAH2 study, 2013–2019.

AURAH2, The Attitudes to and Understanding of Risk of Acquisition of HIV 2; CI, confidence interval; GBMSM, gay, bisexual, and other men who have sex with men; PY, person-years.

https://doi.org/10.1371/journal.pmed.1003677.g001

The most common age category at the time of new HIV diagnoses was between 35 and 44 years, with a total of 13 men (39.4%) in this age range being diagnosed with HIV, followed by men in the age category of 25 to 34 years (8 men, 24.2%), <25 years (7 men, 21.2%), and ≥45 years (5 men, 15.2%). The median (IQR) age at time of new HIV diagnosis was 35 years (26 to 40).

Association of baseline factors with incident HIV

Table 1 presents the association of baseline factors with incident HIV diagnosis. In univariable mixed-effects Weibull proportional hazard models, the factor most strongly associated with HIV acquisition was reporting injection drug use in the past 3 months, with an almost 28-fold higher rate compared to men who did not report recreational drug use (HR 27.96, 95% CI 6.99 to 111.85, global p < 0.001). The HIV IR among people who injected drugs was 4.74 (95% CI 2.13 to 10.54) per 100 PY. Having used at least 1 noninjection chemsex-related drug was also strongly associated with HIV acquisition (HR 6.45, 95% CI 1.84 to 22.64, compared to no drug use); the association with non-chemsex–related drugs was weaker (HR 3.73, 95% CI 0.99 to 14.05).

Other sexual/HIV-related behaviour risk factors were strongly associated with increased risk of HIV infection: CLS (HR 3.75, 95% CI 1.31 to 10·74, p = 0.014), greater number of CLS partners, with increased risk for those having at least 2 partners (HR for 2 to 4 partners 4.05, 95% CI 1.29 to 12.72; HR for 5 to 10 partners 9.60, 95% CI 2.58 to 35.76, HR for more than 10 partners 14.05, 95% CI 4.11 to 47.98, compared with no CLS, global p < 0.001), CLS with HIV–positive partners (HR 6.45, 95% CI 3.15 to 13.22, p < 0.001), versatile CLS role (HR 6.35, 95% CI 2.18 to 18.51, p < 0.001), group sex (HR 8.81, 95% CI 3.07 to 25.24, p < 0.001), and sex for drugs or money (HR 3.27, 95% CI 1.14 to 9.38, p = 0.027) in the past 3 months; reporting a bacterial STI diagnosis in the past 12 months (HR 3.95, 95% CI 1.81 to 8.63, p = 0.001), reporting more than 10 new sexual partners in the past 12 months (HR for 11 to 49 new partners 3.17, 95% CI 1.39 to 7.26, HR for 50 to 99 new partners 4.40, 95% CI 1.35 to 14.29, HR for 100 or more new partners 4.84, 95% CI 1.05 to 22.41, compared to 0 to 10 new partners, global p = 0.001) and having used PEP in the past 12 months (HR 2.29, 95% CI 1.09 to 4.81, p = 0.029).

For socioeconomic and demographic characteristics, lower level of education was associated with increased risk of HIV infection (HR for no qualification 4.65, 95% CI 1.07 to 20.14 compared to university degree, global p = 0.014). There was some evidence that nonemployed men were at lower risk of infection than employed men (HR 0.16, 95% CI 0.02 to 1.19, p = 0.074).

Adjustment for age at baseline, country of birth and ethnicity, sexual identity, and education did not materially change the associations between incident HIV and baseline factors (S2 Table). There were no significant associations of age group, housing status, financial status, relationship status, HIV test in the past 6 months, fisting or sex toys use in the past 3 months, PrEP use in the past 12 months, smoking status, alcohol consumption, country of birth and ethnicity, sexual identity, year of enrolment, depressive symptoms, and anxiety symptoms at baseline with risk of HIV infection (Table 1).

Prevalence of sexual risk behaviours over time

Fig 2 shows the trends in reported sexual risk behaviours, drug use, and the use of PrEP and PEP by calendar year, based on all available baseline and follow-up questionnaires from all 1,162 participants enrolled (total 4,439 questionnaires). Fig 2A shows that the annual prevalence of CLS with 2 or more partners in the past 3 months increased somewhat from 38.3% to 41.0% (p-value for linear trend from GEE logistic model = 0.006) between 2013/2014 and 2018, while group sex declined substantially from 46.7% to 24.2% (p < 0.001), as did bacterial STI diagnoses in the last 3 months from 20.1% to 9.2% (p < 0.001) between 2015 and 2018. For bacterial STIs, data were not included from the baseline questionnaire (since 2013) as it asked about diagnoses in the previous 12 months rather than in the last 3 months. Any recreational drug use in the past 3 months decreased from 57.2% to 45.5% (p < 0.001), while injection drug use (prevalence around 2%, p = 0.903) and the use of at least 1 chemsex-related drug (prevalence between 20% and 30%, p = 0.232) were stable (Fig 2B). Past 12-month PrEP use increased significantly from 2.22% to 43.4% (p < 0.001); on the other hand, PEP use peaked at 21.9% in 2016, then declined to 7.6% in 2018 (p = 0.07) (Fig 2C).

Fig 2. Annual prevalence of sexual/HIV-related behaviours among GBMSM in the AURAH2 study, 2013–2018*.

*Annual reports of (A) sexual risk behaviours in the previous 3 months, data from all available baseline, 4-monthly, and annual questionnaires (N = 4,439 questionnaires), bacterial STI in the previous 3 months, data from 4-monthly and annual questionnaires (N = 3,277 questionnaires); (B) recreational drug use in the past 3 months, data from baseline and annual questionnaires (N = 2,104 questionnaires); (C) PrEP and PEP use in the last 12 months, data from baseline and annual questionnaires (N = 2,085 questionnaires). AURAH2, The Attitudes to and Understanding of Risk of Acquisition of HIV 2; GBMSM, gay, bisexual, and other men who have sex with men; PEP, postexposure prophylaxis; PrEP, preexposure prophylaxis; STI, sexually transmitted infection.

https://doi.org/10.1371/journal.pmed.1003677.g002

Association of time-updated factors with incidence HIV among men who completed at least one online follow-up questionnaire

Among the 622 men who completed an online follow-up questionnaire, 19 were diagnosed with HIV during the period from the date of completion of their baseline questionnaire until June 30, 2019. With a total of 2,495 PY of follow-up time, the overall HIV IR in this subgroup of men was 0.76 (95% CI 0.49 to 1.19) per 100 PY, similar to the overall IR among all men enrolled in the AURAH2 cohort (0.71, 95% CI 0.51 to 1.00 per 100 PY). Table 3 shows unadjusted and adjusted IRRs from mixed-effects Poisson models for factors associated with HIV incidence among these men (total complete observations 3,821 questionnaires). In this analysis, age, partnership status, sexual/HIV-related behaviours, PrEP and PEP use, and health and lifestyle variables were time updated, whereas ethnicity and country of birth, education, employment, sexual identity, financial status, and housing status were fixed variables that were only asked at baseline questionnaires. Longitudinal factors associated with HIV incidence among these men were quite similar to those among the 1,162 men, in particular, injection drug use (unadjusted IRR 21.67, 95% CI 3.96 to 118.30, p < 0.001), chemsex (3.89, 95% CI 1.35 to 11.22, p = 0.012), CLS with 2 or more partners, versatile CLS role, group sex (all in the previous 3 months), bacterial STI diagnosis (in the previous 12 months at the baseline questionnaire and in the past 3 months at the 4-monthly and annual questionnaires), and calendar year.