It is estimated that there are 780,000 HIV positive cases in China. In 2011, there were 92,940 reported HIV/AIDS cases of which 28.4% could be attributed to injection drug use [1], which has driven the HIV epidemic in China since 1989 [2], [3], [4]. The HIV prevalence among injection drug users (IDU) in China ranged from 6.4% to 71.9% [1], [5], [6]. A recent systematic review estimated that there are 2.9 million IDU in China [7]. Although it is commonly believed that some drugs (e.g. heroin) diminish sexual interest among IDU [8], [9], previous studies have revealed that many male IDU remain sexually active [10], [11], [12], [13], [14], [15]. IDU may engage in syringe sharing and unprotected sex and may hence face dual risks of HIV infection [16]. They form a potential ‘bridge’ transmitting HIV from the higher risk IDU population to the lower risk general population, such as their sex partners who are not IDU [17].

HIV-related risk perception is often measured by assessing one’s perceived likelihood of contracting HIV. It is a core element of prevailing health behavior theories [18] such as the Health Action Process Approach (HAPA) [19], [20] and the Health Belief Model [21], [22]. It is an important concept which can be applied to understand HIV-related behaviors and to develop HIV research and interventions. Theoretically, elevated risk perception of HIV infection should prompt individuals to adopt relevant precautionary measures and/or to refrain from adoption of risk behaviors. Attempts to raise awareness and risk perception are hence common strategies of HIV interventions [23], [24]. Risk perception is one of the most popular HIV-related research topics [25], [26], [27], [28], [29], [30], [31], [32]. Despite the significance, its relationship with risk behaviors has not been well studied, possibly due to conceptual and measurement issues.

Previous studies have reported mixed results of positive, negative and null associations between HIV-related risk perceptions and inconsistent condom use in the last few months [33], [34], [35], [36], intention to use condoms consistently in the future [37], [38], [39], previous behaviors of syringe sharing and syringe sharing in the future [40]. Such discrepancies challenge whether attempts to increase risk perception in HIV interventions are evidence-based. It is therefore important to look into methodological reasons beneath the inconsistencies.

One of the plausible reasons of the inconsistencies may be attributed to the cross-sectional study design being used in most of the aforementioned studies [41], [42], [43], [44], [45], [46], [47], which reported associations instead of causal relationships. Positive associations could be resulted if those who had practiced high risk behaviors (e.g. unprotected sex or syringe sharing) in the past would perceive higher chances of contracting HIV, as compared to those who did not practice high risk behaviors. This is known as the reflective hypothesis [48], which can operate in cross-sectional studies involving retrospectively assessed risk behaviors. In contrast, the motivational hypothesis [49] suggests that those who perceive higher chances in contracting HIV in the future would be motivated to adopt preventive behaviors (e.g. condom use or avoid syringe sharing) in the future in order to avoid contracting HIV, as compared to others. In cross-sectional studies, both mechanisms may operate simultaneously to affect the direction of the association, when the question asked involves previous risk behaviors and an unclear time frame with respect to risk perception (e.g. ‘How likely is it for you to contract HIV?’). The relative strength of the two mechanisms would determine the final direction of the association between HIV-related risk and previous behaviors.

As mixed results have also been observed in studies asked about intention to use condoms or intention to share syringes in the future [35], [36], [40] and those studies using longitudinal study design [50], other methodological problems, such as measurement issues, may co-exist with the aforementioned causality problem. The measures of HIV-related perceptions may or may not be specific to the nature of the risk behavior, such as types of risk behavior (e.g. condom use or syringe sharing) and people (e.g. type of sex partners) involved. The measurement of risk behaviors may either be general (e.g. condom use during sex with any sex partner) or specific (e.g. condoms use during sex with regular or non-regular sex partners). It is hypothesized that the use of general versus specific measures of risk perception and risk behaviors would yield associations of different directions.

The study investigated the prevalence of: 1) inconsistent condom use during sex with any female sex partners in the last six months (previous behavior), 2) intention to use condoms consistently during sex with three types of female sex partners (regular, non-regular and commercial female sex partners) in the next six months, 3) syringe sharing in the last six months (previous behavior), and 4) intention to avoid sharing syringes with others in the next six months totally.

HIV risk perception assessments included: 1) levels of two general unconditional sex-related HIV risk assessment variables (not conditional on any characteristics of risk behaviors), 2) one general drug-use-related HIV risk assessment, 3) three measures of sex-related HIV risk assessment conditional on unprotected sex respectively with three types of female sex partners (regular, non-regular and commercial female sex partners), and 4) one measure of drug-use-related HIV risk assessment conditional on syringe sharing with others (see Measures).

In this study, the reflection hypothesis was firstly tested by investigating the corresponding associations between two general HIV risk assessment variables and two types of previous HIV-related behaviors (i.e. inconsistent condom use with any female sex partner and syringe sharing with others in the last six months). It was hypothesized that those practicing such two types of previous risk behaviors would be more likely than others to reflect on the risk level caused by their previous risk behaviors and would therefore possess higher levels of general unconditional sex-related and unconditional drug-use-related HIV risk perceptions.

The motivational hypotheses were tested by investigating the corresponding associations between the four (three sex-related and one drug-use-related) specific HIV risk perceptions conditional on characteristics of the risk behaviors and the intention to avoid the corresponding risk behavior (condom use in the next six months or total avoidance of syringe sharing in the next six months). It was hypothesized that those with higher levels of specific HIV risk perceptions were more likely than others to be motivated to avoid risk behaviors in the future. They would hence have higher intention to use condoms consistently or to avoid syringe sharing totally in the next six months. The null hypotheses suggest statistically non-significant associations. This is one of the few studies tackling issues on the direction of the associations between HIV-related risk perceptions and risk behaviors and is the only one of the type conducted in China.

Corroborating with the results of previous studies [51], the sampled male IDU were at high risk of sexual transmission of HIV/STD – about 90% of them were sexually active and the majority of whom (>80%) were inconsistent condom users. The claim that the majority of IDU would have their sexual drive suppressed was not supported by our data. The prevalence of syringe-sharing among our participants was however, lower than that of other previous similar studies [52] and was consistent with previous findings obtained from local behavioral surveillance studies (personal communication with local CDC staff). It may be due to the extensive intervention efforts in the study sites, or reporting bias, or other unknown reasons.

The aforementioned observations on a high level of sexual risk but relatively low prevalence of syringe sharing suggested that male IDU might have overlooked the risk of sexual transmission of HIV. They might feel that the risk of HIV transmission via sex would be much lower than that via syringe sharing. Our results support the contention. It is interesting to see that the overall or general risk perception in this study population was low (around 10% felt susceptible for contracting HIV). However, the level of risk perception conditioned on syringe sharing was very high and the levels of risk perceptions conditioned on various types of female sex partner varied and were in between those of general risk perception and syringe sharing. Moreover, whilst harm reduction services such as methadone maintenance treatment targeting IDU in China are commonly available and are relatively effective [53], [54], such interventions might have underemphasized prevention of sexual transmission of HIV. The results reminded us of the importance of condom use promotion in the study population. Integrated services are required to reduce dual risks of HIV transmission faced by IDU in China.

Our findings showed that general risk perception was not significantly associated with previous sexual risk behaviors (inconsistent condom use in the last six months), reinforcing the contention that sexual risk may have been under-emphasized. General risk perception was associated with previous syringe sharing behaviors, supporting the reflection hypothesis - previous risk behaviors might have led to increase in current HIV risk perceptions: ‘As I had had unprotected sex, my chance to contract HIV should be high’.

The results were totally different when behavior-and-partner-specific risk perceptions replaced general measures and behavioral intention on consistent condom use, as well as when behavioral intentions on avoiding syringe sharing totally replaced previous syringe sharing. Such conditional risk perceptions were consistently and strongly associated with behavioral intentions to avoid both sex-related and drug-related risk behaviors in the future. Instead of supporting the aforementioned reflection hypothesis, the results hereby supported the motivational hypothesis – a higher level of conditional risk perception motivates individuals to avoid risk behaviors in the future (‘As the chances of HIV transmission associated with unprotected sex and syringe sharing are high, I would avoid such risk behaviors in the future’).

The results of the aforementioned hypothesis testing have strong implications on research design and interpretation of research results. First, researchers should be reminded that the choice of different types of risk perception measures in HIV-related studies (general or specific) and the choice of time frame (previous behavior or future intention) regarding the measures would have direct and strong impacts onto the results. Cautions are hence required. Furthermore, our results have encouraging implications on the design of HIV prevention services. It is still important to remind male IDU of the high risks involved in drug-related and sex-related behaviors. According to our results, such interventions would increase their intention to adopt HIV-related protective measures in the future.

With the harm reduction programs such as MMT in place, health communication messages about HIV risk associated with syringe sharing seem to have been better received as compared to those associated with sexual risk among male IDU. The risk perception involving unprotected sex with female RP of IDU was especially low. Many female RP of male IDU are IDU and are exposed to high risk of HIV transmission via drug use or sex work [55]. Amongst male IDU, there might be a false sense of security regarding low risk of HIV sexual transmission via female RP.

This may be the only study comparing the relationships between various types of conditional/unconditional HIV-related risk perceptions and various types of risk behaviors/intentions (sex-related behaviors with different types of partners and drug-related behavior) among male IDU. Our novel findings contribute to the understanding of the puzzle regarding previous inconsistent findings. Risk perception is a construct of key health behavioral theories. We contend that the aforementioned discrepancy can be partially attributed to the choice of measures on risk perception (general versus specific perception) and risk behaviors (previous behaviors versus future intention). We infer that risk perception is a useful construct in designing HIV prevention among male IDU.

It is warranted to develop and to refine measurements of conditional HIV risk perception. Researchers should be made aware about the heterogeneity and limitations on the currently used measures. Further research and debates are warranted. Such methodological development is a required step to advance behavioral health theories involving risk perceptions to another level. Its importance is not limited to HIV research. Moreover, though our findings were obtained from male IDU, the results should also be applicable to other groups that are vulnerable to HIV prevention, such as men who have sex with men (MSM). Many MSM in China are having unprotected sex with females. It is interesting to investigate risk perception directed to transmitting HIV from MSM to their female sex partners and the associations between such risk perceptions and condom use with female sex partners.

The study has some limitations. It only included male but not female IDU due to resource limitation. Many female IDU in China are FSW [56] who may possess different HIV-related risk perceptions and associations between such risk perceptions and behaviors or intentions. However, the issues discussed are general concerns in understanding other types of health behaviors in other populations. Another limitation was that most of the measures were self-constructed, though references have been made to previous studies. Few studies compared these measures. Limited by the length of the questionnaire, many measures were single-item indicators. Future studies are required to validate new conditional specific risk perception scales. Moreover, non-probabilistic sampling was performed as there was no sampling frame. Respondent driven sampling could have improved representativeness of the findings though other studies have used our study design [57]. Reporting bias due to social desirability may have been affected the findings [58] though the study was anonymous and privacy was ensured in data collection. As this was a cross-sectional study, we could only compare the scenarios using dependent variables of previous behaviors versus variables on future intentions, but not prospectively with future behaviors. Behavioral intention, a construct of the Theory of Planned Behaviors, is one of the strongest predictor of actual future behaviors [59]. Finally, the sample size of participants having had had sex with female sex workers and non-regular sex partners were relatively small.

In conclusion, our results suggested that the discrepancy of the relationship between HIV risk perception and risk behaviors/intentions may be partially explained by the choice of general versus specific tools for risk perception assessments. Both the reflective hypothesis (e.g. positive associations between risk perception and risk behavior) and the motivational hypothesis (e.g. negative associations between risk perception and future behavioral intention) could be supported by the data obtained from the same study, depending on the choice of measures. The support of the motivational hypothesis involving specific measures and behavioral intentions lends a warranted evidence base to the HIV prevention strategy in elevating HIV-related risk perception. HIV workers should be made aware of these results, which have both important theoretical and service implications. Future longitudinal studies are warranted to construct better risk perception measures, allowing for clearer interpretations. Such risk perception is an important but under-emphasized research area which should go beyond investigating HIV transmission among IDU.