Our study is an ongoing prospective open cohort of VHCT among MSM in Tianjin, China. The VHCT project was conducted at the “Shenlan” community-based organization (CBO) and at venues where MSM gather (e.g., bathhouses, bars) and was implemented by professional staff at the CBO who were also MSM. All participants completed a standard questionnaire during HIV counselling to collect information about their demographics, sexual risk behaviour and history of health services. Each participant underwent a rapid HIV antibody test and an amplified nucleic acid test for HIV confirmation. During the 20 min waiting for the rapid test results, all participants were provided condoms and lubricants as compensation for their cooperation and were provided sex education and advice based on the responses on their questionnaires. The results of the amplified nucleic acid test took 1 week to come in, and all participants were informed of their results. Participants were recommended to return for HCT every 3 months. However, HCT was provided whenever MSM wanted an HIV test due to the high-risk sexual behaviour of this population.
Eligible participants included MSM who were HIV-negative at enrolment and had a history of at least two visits. For individuals living with HIV at follow-up, only MSM diagnosed with AEH infection were included. Of the 6565 MSM enrolled in the study from October 2011 to December 2019, 6133 MSM were excluded from the present study (432 were diagnosed with HIV at enrolment, 4096 had only one visit, and 63 had chronic HIV infection). Finally, a total of 1974 MSM were included (Additional file 3: Fig. S1). The average follow-up time was 1.84 person-years.
Demographic data included age (< 30, 30–49 and ≥ 50), marital status (married or not married), household registration location (local or nonlocal) and time of local residence (≤ 1 year or > 1 year).
The following sexual risk behaviours were assessed: working as a male sex worker (MSW) (yes or no), history of anal sex (yes or no) within the previous 6 months, number of instances of anal sex within the previous 7 days, condom use during last sexual intercourse with a man (yes or no), commercial sexual behaviour (yes or no), multiple sexual partners (yes or no) and number of anal sex partners within the previous 6 months.
If participants accepted HIV risk reduction education or condom/lubricant distribution from trained peers and institutions (e.g., CBO, Centers for Disease Control (CDC) and hospitals) in the last year, they were considered to have received health services. We also assessed whether participants underwent HIV testing in the last year (yes or no).
The sexual risk behaviour score was defined as follows. Four behaviours considered to be associated with significant HIV infection in previous literature [10, 13, 14] were used to construct sexual risk behaviour scores: condom use during last time engaging in anal sex with a man within the previous 6 months, frequency of condom use within the previous 6 months, number of times engaging in anal sex within the previous 7 days and number of sexual partners within the previous 6 months. Variable assignments are displayed in detail in Additional file 1: Table S1. The sexual risk behaviour score was calculated for each participant at each HCT visit. Scores ranged from 0 to 6, and higher scores indicated more risky sexual behaviour.
AEH infection was defined according to the following criteria: (1) HIV-1 RNA level of ≥ 1000 copies/mL for two nucleic acid amplification tests (NAAT, Roche COBAS TAQMAN48) in the following situations : (a) negative fourth-generation Ab/Ag screening test (SB rapid test, HIV Ag/Ab combo, Alere, CFDA registered) result but positive HIV pooled PCR result; (b) positive fourth-generation Ab/Ag screening test result and negative/indeterminate ELISA (Wantai Biological Pharmaceutical Co., Ltd, Beijing, China) test result; (c) negative or indeterminate Western blot (WB, MP Biomedical Asia Pacific Pte Ltd, Singapore) test result; (2) a positive HIV-1 antibody test result and a documented negative HIV-1 antibody test result within the previous 6 months .
Group-based trajectory modelling (GBTM)
Across the follow-up period, group-based trajectory modelling (GBTM) was used to identify sexual behaviour trajectories by sexual risk behaviour score. One of the main advantages of this method is that the missing values are considered; therefore, we can include participants with at least two follow-up visits. The PROC TRAJ SAS procedure was used to calculate the probability of each participant belonging to each trajectory group, and individuals were assigned to trajectories based on their highest probability of trajectory membership. By fitting a series of models with three to five trajectories, several models were constructed assuming linear, quadratic, and cubic shapes of the trajectory group curves. In determining model fit and the optimal number of trajectory groups, model fit statistics and interpretability were considered. Statistics included the Bayesian information criterion (BIC), Akaike information criterion (AIC) and significance of the shape of trajectory group curves. The model fitting process started with a cubic specification for the shape of trajectory group curves, and then, significant terms were assessed [17,18,19]. After excluding nonsignificant terms, the trajectory models were refitted again. The models used a zero-inflated Poisson distribution to account for the large number of participants who reported having no risky sexual behaviours.
Logistic regression and generalized estimating equation (GEE)
After the optimal trajectory model was selected, logistic regression was used to generate the AEH infection risk of different sexual behaviour trajectories. A generalized estimating equation (GEE) was constructed to compare the risk factors for sexual behaviour trajectories. The multivariable analysis included variables with 2-tailed \(P\) < 0.1 in the univariate analysis.
Restricted cubic splines (RCS)
The association between the follow-up period and the risk of AEH infection was evaluated with restricted cubic splines (RCSs) based on the Cox proportional hazards model. Five knots were chosen at the 5th, 25th, 50th, 75th, and 95th percentiles of follow-up time. SAS 9.4 (SAS Institute Inc.) was used for statistical analyses.