Comparing the impact of sexualised drug use with and without chemsex on sexual behaviours among men who have sex with men in China: a national multi-site cross-sectional study

Study population

Measures of SDU and chemsex

Chemsex was defined as the lifetime use of drugs for sexual purposes, including poppers,^9,19,20 crystal meth (also known as ‘Ice’),^19,21 ketamine,^19,22 ecstasy (MDMA),^19,22 or GHB/GBL^19,22 in conjunction with sexual activities, categorising participants into the ‘chemsex’ group (see Supplementary material Table S1). Participants who reported a lifetime use of alcohol,^1,19,23 erectile dysfunction medication (e.g. Viagra),^1,24 marijuana,^1,24 or heroin^1,24 during sexual activity but did not use chemsex drugs were categorised as the ‘SDU without chemsex’ group (Table S1). Conversely, individuals who did not report SDU were categorised as the ‘non-SDU’ group.

Measures of risk of HIV/STIs

To analyse the association between SDU and the risk of HIV/STIs infection, we used the MySTIRisk tool²⁵ to calculate the HIV/STIs risk score for each participant. MySTIRisk was a prediction tool trained based on the data from the Melbourne Sexual Health Centre (MSHC) in Australia. The tool employed machine learning techniques to generate reliable estimates of the probability of four types of infections, including syphilis, gonorrhoea (NG), chlamydia (CT), and HIV.

The MySTIRisk assessment tool was incorporated into our questionnaire designed for Chinese MSM. The majority of the MySTIRisk questions were directly compatible with our study questionnaire. The MySTIRisk tool calculates the risk scores based on the users’ responses to a series of risk-assessment questions. The model outputs percentage scores ranged from 0 to 1. A higher percentage indicates a greater risk.

Measures of sexual behaviour

We estimated the frequency of sexual acts over the past 12 months for each participant. Participants were asked to recall their most recent sexual encounters and report on 10 specific types of sexual acts they engaged in with male partners. The 10 sexual acts were: (1) French kissing; (2) self-masturbating; (3) masturbating another man; (4) masturbated by a man; (5) receptive oral sex; (6) insertive oral sex; (7) receptive rimming; (8) insertive rimming; (9) receptive anal sex; and (10) insertive anal sex. We assumed that the frequency of these acts remained consistent for each participant, and we calculated the occurrence of each sexual behaviour over the past 12 months.

We also analysed the diversity of sexual behavioural patterns during the most recent sexual encounters using the Shannon diversity index, a metric commonly used in microbiome analysis.^26,27 This index helped us characterise the richness and variability of sexual behaviours among participants. We then compared the diversity of sexual behaviours across the ‘chemsex’ group, the ‘SDU’ group, and the ‘non-SDU’ group.

The questionnaire included a question to ask participants the order when the sexual acts occurred in the most recent sexual encounter. Participants were asked to list up to 10 sexual acts in the order they occurred. We defined sexual act pairs as two consecutive sexual acts within this sequence, enabling us to study the patterns of sexual behaviour during a single encounter. By analysing these sexual act pairs, we compared the correlations of chemsex or SDU on sexual behaviour patterns.

To investigate whether SDU affects sexual activity, we calculated the frequency of different sexual behaviours during the last sexual encounter for both the ‘SDU’ and ‘non-SDU’ groups. We then used χ² analysis to determine whether SDU significantly influences the patterns of sexual behaviour. To control for potential confounding factors, we applied the multivariable logistic regression models that included all socio-demographic variables alongside the independent variables. To evaluate model fit and examine the impact of potential confounders, we compared the univariable and multivariable models using the Akaike Information Criterion,²⁸ Bayesian Information Criterion,²⁹ and χ² difference tests. For each model comparison, we computed χ² statistics and corresponding P-values.

Measures of PrEP use

PrEP-naïve participants were those who responded ‘No’ to the question, ‘Have you ever taken PrEP?’. Initiated PrEP participants were those who responded ‘Yes’ to the question, ‘Have you ever taken PrEP?’. For those that initiated PrEP, we further divided the responses into two groups: (1) ‘currently using PrEP’; and (2) ‘stopped using PrEP’. The ‘currently using PrEP’ group answered ‘Yes’ to ‘Are you taking PrEP right now?’. PrEP-naïve participants were further divided into those ‘Interested in PrEP’ and those ‘Not interested in PrEP’.

Statistical analysis

A single-factor ANOVA was used to analyse the risk prediction outcomes, assessing the magnitude of STI risk variables associated with different patterns of SDU. The ANOVA results were evaluated for statistical significance across all groups, ensuring that the analysis captures any differences. Non-parametric tests were used when data violated the homogeneity of variance assumption or did not conform to a normal distribution during the ANOVA process.³⁰ The chi-squared test was employed to perform intergroup statistical comparisons for categorical data. A P-value of 0.05 or less was considered statistically significant. All statistical analyses were performed using Prism 10, SPSS 26, and Python 3.9.

Ethical statement

Permission to conduct the project was obtained from the Human Research Ethics Committee of Zhuhai Centre for Disease Control and Prevention (Ethics documents ID No. [2022] 11).

Socio-demographic characteristics of study participants

In total, 1034 MSM completed the questionnaires. We excluded 142 because the participants were less than 18 years of age, two questionnaires were duplicates, and 94 had no data reported on sexual behaviour. Thus, we included 796 MSM in our final analysis (Fig. 1).

Fig. 1.

Flowchart summarising study design and patient recruitment.

The demographic characteristics of participants are in Table 1. Of these, 571 participants were classified as non-SDU, while the remaining 225 belonged to the ‘SDU’ groups. Subsequently, 68 participants were categorised into the ‘chemsex’ group, while 157 were classified as SDU without chemsex drugs. Within the ‘chemsex’ group, poppers (n = 67, 8.4%) were the most used drug, followed by ketamine (n = 1, 0.1%) (Table S1). For the ‘SDU without chemsex’ group, most used alcohol (n = 132, 16.6%), followed by erectile dysfunction medication (n = 30, 3.8%), and marijuana (n = 1, 0.1%). Among all participants who use erectile dysfunction medication, 60% were young individuals aged 30 years and under.

Age of the participants ranged from 18 years to 60 years, with a mean of 28.2 years (s.d. = 7.43). No significant differences in age distribution were observed between groups (P = 0.467). Participants mostly identified as single or unmarried (n = 712, 89.4%) and as gay or homosexual (n = 612, 76.9%). The ‘chemsex’ group reported a higher frequency (2.9%) of heterosexual orientation compared to other groups (P = 0.010). Notably, 78% of participants voluntarily disclosed their sexual orientation (n = 623). Educational attainment was predominantly undergraduate and above (n = 587, 73.7%). The largest proportion of participants reported a monthly income ranging from RMB5001 to RMB8000 (n = 213, 26.8%).

SDU and HIV/STIs risk scores

A scatter boxplot of the HIV/STIs risk scores among the ‘SDU’ and ‘non-SDU’ groups shows that the median risk scores were similar among the three groups, with the quartiles of the risk score for SDU with chemsex drugs being higher (Fig. S1). Fewer individuals with high-risk scores were in the ‘SDU without chemsex’ group. We did not identify statistically significant differences among the three groups regarding their HIV/STIs risk scores (Kruskal–Wallis test, P = 0.308).

SDU and number of sexual partners

Fig. S2 shows the number of sexual partners in the past 6 months among participants, categorised by their drug use. The distribution of sexual partners across different groups did not follow a normal distribution, as indicated by the P-value from the normality test (P < 0.0001). Therefore, the Kruskal–Wallis test was conducted and showed the group differences (P = 0.004). MSM with SDU without chemsex had a statistically significant difference in the number of sexual partners compared to MSM with SDU with chemsex (P = 0.004). The number of sexual partners in the ‘chemsex’ group increased and was concentrated at a higher level. In contrast, the number of sexual partners in the ‘SDU without chemsex’ group was generally lower, and the extreme values of sexual partners were even rarer. The number of people in different groups with five or more sexual partners was compared in Table 2, and the χ²-test results indicate statistical differences among the groups (P = 0.039). The ‘chemsex’ group (n = 29, 42.7%) had the highest proportion of participants with five or more sexual partners.

SDU and sexual behaviour patterns

We conducted a multivariable logistic regression analysis that adjusted for potential confounders (age, level of education, monthly income, etc.) and compared it with univariable logistic regression models. None of the P-values across group comparisons reached statistical significance (P > 0.05) (Tables S2 and S3). This indicates that potential confounders did not have a significant effect on the comparisons of sexual acts between groups, validating the robustness of our univariable analysis results.

Table 2 shows that the difference in self-masturbation (P = 0.003) and being masturbated by a man (P = 0.015) among the three groups were statistically significantly different. The Shannon diversity index was used to calculate the richness of participants’ most recent sexual activity. The non-parametric statistical test results of the Shannon diversity index under different group conditions (Fig. 2). There was a statistically significant difference between the SDU with chemsex drugs and the ‘SDU without chemsex’ groups (P = 0.030). The one-way ANOVA test indicated a statistically significant difference between the diversity of sexual activity and SDU (P = 0.031).

Fig. 2.

Diversity of sexual behaviours among the three groups.

Fig. 3 is a radar plot that compared the number of sexual activities in the past 12 months among groups. Kruskal–Wallis test results showed that there were group differences in masturbating another man (P = 0.050). In general, the ‘SDU with chemsex’ group engaged in most types of sexual behaviours more frequently, followed by the ‘SDU without chemsex’ group. For the ‘non-SDU’ group, self-masturbation was most frequent, and they engaged in other sexual behaviours less frequently than the other two groups.

Fig. 3.

Radar plot of the number of sexual acts in the past 12 months among the three groups.

We further compared the differences in sexual behaviour pairs among different groups. The ‘chemsex’ group were more likely to be involved in sexual activity pairs with receptive or insertive anal sex. Individuals in the ‘chemsex’ group had a major proportion of those who engaged in being masturbated by a man first followed by receptive rimming compared to the other two groups (4.4%, P = 0.027) (Table 2). Participants in the ‘chemsex’ group had a lower proportion who engaged in insertive oral sex followed by receptive oral sex (1.5%, P = 0.050). Individuals in the ‘SDU without chemsex’ group had the highest proportion of insertive oral sex followed by receptive oral sex (10.2%, P = 0.015). After engaging in receptive oral sex, the ‘chemsex’ group was more likely to continue with self-masturbation (n = 4, 5.9%) or receptive rimming (n = 15, 22.1%) but showed a lower likelihood of engaging in insertive anal sex (n = 5, 7.4%). For statistical results for all sexual activities and sexual activity pairs, see Table S4.

We compared the correlation of SDU on different sexual behaviours and counted the number of occurrences of specific sexual behaviours separately according to whether they use other drugs for sexual purposes or not (Table S4). Kissing, as typical sexual behaviour, occurred more frequently in the ‘non-SDU’ group than in the other groups. Most (82.7%) of the ‘non-SDU’ group had kissed during their last sexual encounter. In addition to kissing, the three most common sexual behaviours in the ‘chemsex’ group were receptive oral sex (n = 51, 75.0%), insertive oral sex (n = 37, 54.4%), and receptive anal sex (n = 36, 52.9%).

SDU and PrEP use

To investigate the influence of SDU on PrEP use, the variations among groups were counted based on the distinct participant attitudes regarding PrEP across the four groups (Table 2). The data show that men who did not use drugs for sexual purposes had a higher percentage of PrEP use (n = 58, 10.2%), while men who used SDU without chemsex drugs were more likely to have no interest in PrEP (n = 59, 37.6%). More people in the ‘SDU with chemsex’ group chose either to stop using PrEP (n = 9, 13.2%) or expressed interest in PrEP (n = 32, 47.1%). However, the differences in participant numbers among these three groups did not reach statistical significance (P > 0.05).