Risk factors associated with HIV infection among men who have sex with men in China: a meta-analysis

Background: Although human immunodeficiency virus (HIV) prevalence in the general population is very low in China, high infection rate has been reported among men who have sex with men (MSM). We conducted a meta-analysis to identify HIV infection associated risk factors among MSM in China, thus we can further understand the high-risk population and provided basic information to further develop specific and effective interventions of HIV prevention. Methods: A comprehensive literature search was conducted in several public databases, the relevant articles which published from January 2010 to June 2018 were identified, and a meta-analysis was performed according to these included studies. The odds ratio (OR) and its 95 % confidence intervals (CI) of each risk factor among MSM in China were pooled by using a random-effects model or fixed-effects model when appropriate. Results: A total of 23 articles were included and analyzed. The pooled results revealed that non-local residency (OR=2.31, 95% CI: 1.05, 5.08), education less than junior high school (OR=1.73, 95% CI: 1.36, 2.21), engaging in commercial sex (OR=2.99, 95% CI: 1.02, 8.72), preferred receptive sexual role (OR=2.43, 95% CI: 2.09, 2.83), having anal bleeding during anal intercourse (OR=2.22, 95% CI: 1.60, 3.07), having no HIV test in the last 12 months (OR=2.17, 95% CI: 1.45, 3.25), having unprotected anal intercourse (UAI) in the last 6 months (OR=2.06, 95% CI: 1.69, 2.50), recreational drugs use (OR=1.90, 95% CI: 1.53, 2.36), preferred versatile sexual role (OR=1.69, 95% CI: 1.35, 2.21), inadequate HIV related knowledge (OR=1.63, 95% CI: 1.26, 2.11), having multiple sexual partners (MSP) in the last 6 months (OR=1.35, 95% CI: 1.24, 2.47), having infection of syphilis

Conclusions: Continuous education and further interventions such as Pre-Exposure Prophylaxis (PrEP) should be prioritized for those MSM who engaged in high-risk behaviors. Background Acquired immune deficiency syndrome (AIDS), caused by infection with human immunodeficiency virus (HIV), remains a significant public health problem with an estimated 1.1 million death (1) and more than 1.5 million newly identified infections each year globally (2). In China, more than one hundred thousand people were newly identified with HIV infection annually in the past few years and the number continued to grow (3)(4)(5)(6)(7)(8)(9).
Although HIV prevalence in the general population is very low (0.055% in 2017) (10), high infection rate has been reported among men who have sex with men(MSM) in China (11).
Previous studies reported the prevalence increased significantly from 0.9% in 2003 to 7.3% in 2017, peaking at 8.0% in 2015 among MSM in China (10)(11)(12)(13) Fig.1a.Meanwhile, with the changing of transmission mode from primarily blood-born spreading to sexual contract (11,14), the proportion of transmission among homosexuals has continued to gain ground among all groups, rising from 0.3% before 2005 to 28  (ii)quantitative epidemiological studies, including observational studies which conducted among MSM in China; (iii)studies measured the HIV infection associated behavior or factors, and the observed association in standardized form of odds ratio (OR) or risk ratio (RR). In this paper, OR stands for these risk estimates; (iv) studies had sufficient discussion of results for proper data abstraction.
Exclusion criteria. Studies were excluded from this paper if any of the following criteria apply: (i)review articles, editorials, newspapers, modelling studies, case reports and conference report; (ii)studies that specified only one subpopulation, such as students, sex workers, recreational drug users, etc., and only one HIV infection associated factor, such as unprotected anal intercourse(UAI), inadequate HIV knowledge, multiple sexual partners(MSP, defined as having more than two male sexual partners), etc., (iii)studies were lack of OR, RR value and related confidence intervals(CI) for risk estimates.

Quality Assessment
The Agency for Healthcare Research and Quality (AHRQ) an 11-item checklist tool was used to assess bias in cross-sectional or prevalence studies (17). An item would be scored '0' if it was answered 'NO' or 'UNCLEAR'; if it was answered 'YES', then the item scored '1'. Study quality was assessed as follows: low quality (0-3); moderate quality (4-7); high quality (8)(9)(10)(11). The Newcastle-Ottawa Quality Assessment Scale(NOS) tool was used for quality assessment in cohort studies and case control studies (18,19). Studies were judged with 8 questions on three broad perspectives by a 'star system': the selection of the study groups; the comparability of the groups; and the ascertainment of either the exposure or outcome of interest for case-control or cohort studies respectively. Study quality was assessed as follows: low quality (0-3 star); moderate quality (4-6 star); high quality (7-9 star).

Study Selection and Data Abstraction
Two reviewers (J. P. and Y. R.) screened all retrieved articles and completed study selection and data abstraction independently against the eligibility criteria. Data information was extracted from all eligible studies and entered into a standardized excel spreadsheet, included: first author, publication year, study year, location, study design, sampling methods, sample size, age range, HIV prevalence, HIV incidence, education, migration, HIV test, syphilis status, sexually transmitted infections(STI) status, HIV knowledge, UAI, anal bleeding during anal intercourse, preferred sexual roles, number of male sexual partners, sex with female partners, recreational drugs use behavior and commercial sex behavior. Not all variables were included in each study, discrepancies and contradictions were resolved by discussions.

Data Analysis
All the meta-analyses were performed using Review Manager 5 statistical software (RevMan 5, Cochrane Community, London, United Kingdom). The pooled OR and 95% CI for each risk factor presented in the form of forest plots. Statistical significance was defined as two-tailed α<0.05. Given prevalence was <10%, the prevalence ratio was used as a proxy for OR given that for rare outcomes, RR approximates an OR (20). Two studies (21,22) reported HIV infection related factors with RR which were conversed to OR directly.
Heterogeneity among stratified group was primarily assessed by the Chi squared (χ2) based Cochran Q statistic (p < 0.10, statistical significant was considered) and I squared (I 2 ) test (13). I 2 values (from 0 to 100%) of 25%, 50% and 75% represented low, moderate and high heterogeneity, respectively (23). If heterogeneity was high (I 2 ≥75%) and significant (p value < 0.10 using Q test), the random-effect model was used for summary risk estimate. Otherwise, the fixed-effect model was used (24,25). Sensitivity analyses were conducted to assess the impact of excluding studies which had high heterogeneity or low quality.

Search results
Overall, 1891 potentially related studies were identified from abovementioned database.
After de-duplication, 1489 articles were screened in titles and abstracts and 1351 articles were excluded per our selection criteria. Then, 138 articles were reviewed in full. Finally, a total of 23 articles were included and analyzed. The PRISMA flow diagram for included studies showed in Fig.2.

Characteristics of Included Studies
The summary information of studies included in this meta-analysis and evaluation of study quality is shown in Table 1. Of the 23 included articles, 14 articles were published in Chinese (21,(26)(27)(28)(29)(30)(31)(32)(33)(34)(35)(36)(37)(38) and 9 in English (22,(39)(40)(41)(42)(43)(44)(45)(46). These studies mainly conducted from 2008 to 2015 and the sample size ranged from 250 to 47,231. The age of these participants ranged from 16 to 85 years. Most of studies recruited study participants utilized a combination of recruitment methods. All the included studies were observational studies. Nineteen were cross-sectional studies, 3 were cohort studies and one was casecontrol study. Twenty studies reported the HIV prevalence and two cohort studies reported the HIV incidence. For consistency, one cohort study (42) was treated as cross-sectional data, with reporting on baseline data. Thirteen out of twenty-three included studies met the criteria of "high quality" and ten studies met "moderate".  Table 2 shows the summary of the meta-analysis for demographic and behavioral factors related to HIV infection. Using the random effects model, the risk factors include: having UAI in the last 6 months, commercial sex behavior, having sex with female partners in the last 6 months, non-local residency (defined as not having permission for permanent residence), having no HIV test in the last 12 months.

Risk factors associated with HIV infection
Using the fixed effects model, factors associated with HIV infection are: having MSP in the last 6 months, recreational drugs use behavior, education less than junior high school,

Sensitivity analysis
There were significant heterogeneities of results across studies in effect analysis of having UAI in the last 6 months (I 2 = 79%, p<0.00001), engaging in commercial sex(I 2 = 84%, p = 0.002), having sex with female partners in the last 6 months (I 2 = 82%, p = 0.0007), non-local residency (I 2 = 99%, p<0.00001) and having no HIV test in the last 12 months (I 2 = 94%, p<0.00001). However, sensitivity tests did not reveal any individual study that exerted impact on the overall estimate.

HIV Prevalence
The HIV prevalence among MSM in China has increased continuously for the last decade.

Risk factors associated with HIV infection
The results of this study suggest that having UAI is a major risk factor for HIV infection.
On average, 49.32% (95% CI: 41.17%, 57.48%) Chinese MSM have engaged in UAI among our included studies. This is consistent with the previous meta-analysis (47) that pooled prevalence rates of UAI with any male partner was 53%(95% CI: 51%, 56%). One study Therefore, other effective intervention methods besides education, promotion and condom distribution, should be considered.
Recreational drugs such as rush poppers, ketamine and methamphetamine, has quickly emerged and increasingly become popular in recent years among MSM (49). The pattern of drug use among Chinese MSM population has changed from using traditional drugs (such as opium, heroin) to the abuse of synthetic drugs (such as methamphetamine, ketamine), and towards the newly emerged drugs (e.g., rush poppers) (50,51). Several studies' results revealed that rush poppers were the most popular recreational drug among MSM in many areas over the last several years(51-54). As newly emerged drugs, rush poppers are not defined as an illicit drug in China and are easy to access with low price from internet (54,55). Recreational drug users can have enhanced feelings of sexual desire, stamina and intoxicating highs (50). Under the effect of recreational drugs, drug users are more likely to have sexual disinhibition and engage in high risk behaviors, such as having multiple sexual partner (56), participating in group sex (57), engaging in UAI and commercial sex (52,58), which may further facilitate HIV transmission. In our study, pooled results showed that recreational drug users had a 1.9 times higher risk of HIV infection than nonusers. Interestingly, one study found that recreational drug users were reported as more likely to have HIV testing experience than non-users (58), perhaps due to increased anxiety over HIV infection post exposure. Another study found MSM with high education level were more likely to get recreational drugs (59). One possible factor could be that better educated MSM may be better at using internet to acquire both information and source of synthetic drugs.  (60,62,67). In that case, the person who is easily infected through URAI could have greater possibility of transmission through unprotected insertive anal intercourse (UIAI) (68).
Have syphilis infection which was reported as a risk factor related to HIV infection by most studies (n = 17) in our review. The relationship between HIV and syphilis has been extensively studied (69)(70)(71). Some studies' results revealed that syphilis infection would act as a cofactor for HIV infection (72)(73)(74)(75) On one hand, some studies have suggested that those MSM who had lower education level and lack of HIV related knowledge such as HIV transmission routes were more likely to engage in UAI (80)(81)(82). And these two factors were also positively associated with HIV infection revealed in our study. Obviously, for these population, publicity and education are needed but the content should include comprehensive sex education and not be limited to HIV related knowledge. One recent survey indicated that, even in the college, only half of the respondents received school-based sex education before (83). This suggests that comprehensive sex education need to cover the college population as well, which should include HIV/STI related knowledge, usage of condom, sexual orientation and safe sex, etc. The approach should also be diverse to include peer education, community education and online education through apps, etc.
On the other hand, researches also shown that separation of knowledge and behavior is quite common among MSM (84)(85)(86), that is, despite high level of knowledge and awareness of HIV and prevention methods, many still act in contrary to what they know. For this population, having high risk behaviors can't be attributed to lack of related knowledge, so

Limitations
There are some limitations in this study. Firstly, our results are based on most of cross-sectional data, which provide HIV prevalence. We did not attempt to have meta-regression for HIV prevalence. Because the objective of this study was not to obtain a single summarized HIV prevalence and most of included studies were conducted in large and medium sized Chinese cities at different study year. The pooled HIV prevalence may not objectively represent HIV prevalence among MSM in China. Secondly, the heterogeneities of included studies are also cause of different sample size, sampling method, study year, etc. For example, 12 studies were conducted in a single city in China with small sample size, 10 studies were conducted in more than one city and one is a national-wide study among 61 cities with large sample size. Moreover, the data collected on demographic factors were insufficient like age and income, due to limited raw data provided in studies and the inconsistency of range intervals for age and income. Further studies into how the relationship between age, income and HIV should be considered. Lastly, though publication bias was not formally assessed in our study, as analytical methods to test for publication bias, such as funnel plots and funnel plot asymmetry tests, may not be appropriate for observational data(98). But the possibility of publication bias could not be excluded.   Supplementary Materials S1b.tif