Gay, bisexual, and other men who have sex with men (MSM) account for over 60% of all new human immunodeficiency virus (HIV) cases annually in the United States.1 Although prevention efforts originally relied on frequent testing and early treatment of HIV, the prophylactic use of antiretroviral drugs now shows promise in decreasing HIV burden in this population.
In 2012, the US Food and Drug Administration approved tenofovir disoproxil-emtricitabine (Truvada), an antiretroviral medication, as preexposure prophylaxis (PrEP), for daily dosing by high-risk HIV-negative individuals to prevent HIV infection. The US Centers for Disease Control and Prevention (CDC) state that PrEP is appropriate for “[a] gay or bisexual man who has had anal sex without a condom or [has] been diagnosed with a (sexually transmitted infection [STI]) in the past 6 months.”2 The CDC also recommends PrEP for HIV-negative MSM who are in a relationship with an HIV-positive partner. Although this recommendation is useful, it omits important factors that may contribute to the risk for HIV infection, such as substance use, frequency of sexual contact, and other predictors shown to be related to HIV infection.3,4
In a longitudinal analysis of HIV risk and infection among initially HIV-negative MSM, Menza et al.3 found that initially HIV-negative MSM who eventually tested HIV-positive were more likely at baseline to have a history of STIs, used inhaled nitrates or methamphetamines, engaged in condomless anal intercourse (CAI) with an individual of HIV-positive or unknown status, and to have 10 or more partners in the past year, compared with individuals who remained HIV-negative. In a subsequent study, Smith et al.4 found that HIV seroconverters were significantly more likely than non-seroconverters to be younger, have a higher number of sex partners, a higher proportion of HIV-positive partners, greater number of CAI encounters, and report use of either inhaled nitrates or methamphetamine. While current PrEP guidelines consider STI history and CAI in determining PrEP candidacy, other predictors of HIV infection including substance use and number of sex partners are not considered.
Syndemics Theory proposes that psychosocial factors should be considered in a comprehensive HIV risk assessment. Briefly, Syndemics Theory proposes that HIV is not 1 epidemic but instead is a complex web of overlapping epidemics of sexual risk behavior, substance use, and psychosocial conditions.5 Each of these epidemics promote and reinforce each other to determine individual and population level HIV disease burden. Although the previous analyses of longitudinal HIV risk considered sexual risk and substance use, subsequent analyses have supported inclusion of additional psychosocial components.6–8
Guided by Syndemics Theory, the primary aim of this retrospective study is to use behavioral and HIV testing data from a racially diverse sample of HIV-negative MSM to determine the significant variables at baseline predictive of HIV infection at follow-up and subsequently use these findings to construct an HIV risk score to inform PrEP candidacy among MSM. The secondary aim is to compare the fit of the current model to the current CDC PrEP criteria and to HIV risk models proposed by previous studies.
MATERIALS AND METHODS
The Los Angeles LGBT Center (The Center) is a federally qualified health center that provides primary health care, HIV specialty care, and HIV/STI testing to over 20,000 unique clients annually.
Data were collected for all MSM who received HIV testing services from January 2009 to June 2014. The HIV/STI testing counselors conducted face-to-face behavioral risk assessments that included questions on demographics, STI history, behavior during the last sexual encounter, and substance use.
After the behavioral risk assessment, blood was obtained for HIV testing using the OraQuick ADVANCE Rapid HIV-1/2 Antibody Test (OraSure Technologies, Inc., Bethlehem, PA). Provided the initial antibody test was positive, a second confirmatory Uni-Gold Recombigen HIV-1/2 antibody test (Trinity Biotech, Wicklow, Ireland) was performed to confirm HIV infection. The HIV nucleic acid amplification testing was performed from remnant specimens of those individuals whose rapid HIV test was negative or individuals whose first antibody test was positive but second antibody test was either negative or inconclusive. Individuals electing to also receive STI testing were instructed to self-collect urine and rectal samples for Neisseria gonorrhoeae and Chlamydia trachomatis. A lab assistant collected a throat swab for pharyngeal N. gonorrhoeae testing and obtained additional blood to test for Treponema pallidum (syphilis) via rapid plasma reagin testing.
Individuals are included in the analysis if they (1) identified as MSM, (2) tested HIV-negative at their baseline visit, (3) reported sex with another man in the year prior to their baseline visit, (4) tested for HIV on at least 1 occasion after their baseline visit during the analysis period, and (5) had a conclusive HIV result during their final testing visit. Men who have sex with men are defined as individuals who at their baseline visit reported a biological sex and gender identity of male and identified either as gay or bisexual or reported having sex with another man in the past year.
Syndemics Theory specifies 3 constructs (biological, behavioral, and psychosocial) that influence one another and ultimately influence HIV incidence. Each of those domains was represented by questions included within the standard risk assessment. Due to documented differences in HIV incidence by both age group9 and race/ethnicity,9–11 these demographic covariates are also included in the predictive model.
All predictors used were assessed at baseline. We tested each predictor individually using bivariate Cox proportional hazards models for continuous predictors and bivariate Kaplan-Meier estimators with log-rank tests for categorical predictors. We checked the proportional hazards assumption for predictor variables with the Kolmogorov-type Supremum Test,12 functional form using Martingale residuals, and overall model adequacy (calibration) with the Grønnesby and Borgan test.13 We constructed a multivariable survival model by operationalizing the constructs outlined by Syndemics Theory given the variables available. Neither backward elimination nor forward selection was implemented.
To create the HIV-risk algorithm, continuous variables were categorized. Number of sexual partners in the last three months was dichotomized as ≤3 and >3, where 3 was the median value. For age difference of the last sexual partner, individuals whose age was within 5 years of the client age were classified as being of similar age, individuals more than 5 years older as “older,” and individuals more than 5 years younger as “younger.”
All categorical predictors were then arranged so the reference group was the lowest risk group. We refit the multivariable model with the original categorical predictors and the new categorical predictors for number of sexual partners and age of last sexual partner. Coefficients were added for each individual in the data set who had values for all predictors. The sum was then exponentiated to create a risk score that compares that person's hazard of HIV infection to the hazard of a hypothetical person in the lowest risk group.14–16 After rounding to the nearest integer, the final range of risk scores was between 1 and 74. The proportion of HIV-positives and HIV-negatives with risk scores greater than or equal to each risk score cutpoint were tabulated.
To assess model discrimination, we applied the techniques outlined by Harrell et al.17 by first sampling 9481 patients with replacement from the original sample (bootstrap sample). We fit single predictor survival analyses on the bootstrap sample for each of the 18 predictors available that were not deemed to be multi-collinear. Predictors significant at P = 0.05 were combined into a bootstrap multivariable survival model (bootstrap full model). The bootstrap full model was fit on the bootstrap sample. Records not included in the original bootstrap sample formed the holdout sample. The C-statistic was calculated on the holdout sample, and this process was repeated 100 times. The 100 C-statistics were then averaged to measure predictive accuracy of the model (Harrell's C-index).
Lastly, we evaluated the extent to which our model compares with previously published models from Menza et al.,3 Smith et al.,4 and the CDC criteria for PrEP.2 Specifically, we selected variables from our data set most closely approximating measures described in these publications, and then used these variables to refit all of the hazard models in our data set. Therefore, variables from the original studies were adapted as closely as possible using variables from this data set and subsequently analyzed using these data.
The same sample from the multivariable model (n = 8898) was used to test the 4 models. Some variables used in other models were either not available or could not be exactly replicated with our data set. Akaike Information Criterion (AIC) and Schwarz Bayesian (Information) Criterion (BIC) were used to compare the models; lower values indicate better fit with the data.18 The CDC criteria were compared to our model using an F test. All analyses were performed using SAS Version 9.4 (Cary, NC).
Between January 2009 and June 2014, a total of 9481 unique MSM tested HIV-negative at baseline and had at least one subsequent HIV testing visit in the analysis period with a conclusive result. There were 370 HIV infections over a period of 16,894 person-years for an HIV positivity rate of 2.18 infections per 100 person-years. The mean amount of follow-up time for all subjects was 651 days (SD = 497 days). The majority of individuals had less than one year of follow-up time (37%), followed by between 1 and 2 years (25%), 2 and 3 years (18%), 3 and 4 years (10%), and more than 4 years (9%). The number of HIV testing visits over the 5.5 years of follow-up did not differ between MSM who were positive and negative at the end of follow-up (median = 3; mean = 4).
Hispanic and African-American MSM were more likely to seroconvert over time compared to White individuals (Table 1). Younger individuals at baseline were also more likely to seroconvert over time. There were no statistical differences between seroconverters and nonseroconverters by sexual orientation.
Individuals who reported a history of chlamydia, gonorrhea, and/or syphilis were more likely to seroconvert when compared to individuals who reported no reported history of these STIs (Table 2). In addition, individuals who tested positive for any of these STIs at baseline were more likely to seroconvert at follow-up.
Seroconverters were more likely than non-seroconverters to report not using condoms during last receptive anal sex and to report more sexual partners in the last 30 days (mean = 3.2 sex partners for HIV-positives vs 2.5 sex partners for HIV-negatives) and the last three months (6.8 sex partners for HIV-positives vs 4.9 sex partners for HIV-negatives) (Table 3). Seroconverters were also more likely to report their last sex partner was the same race/ethnicity compared to nonseroconverters and to report an older partner at last sex. For the psychosocial construct, individuals who reported intimate partner violence (IPV) were more likely to seroconvert compared to individuals who did not report IPV.
Seroconverters were more likely to report past year use of ecstasy, methamphetamine, and inhaled nitrates (Table 4) compared with nonseroconverters. However, HIV seroconversion was not associated with nonprescription use of erectile dysfunction drugs, cocaine, or alcohol use before sex.
The multivariable model showed that Hispanic and African-American MSM were more likely to seroconvert compared to White MSM (Table 5). Men who have sex with men who reported a history of chlamydia, gonorrhea, and/or syphilis in the past year were more likely to seroconvert compared to individuals who never reported these infections. Men who have sex with men who reported receptive anal sex at the last sexual encounter, either with condoms or without, were more likely to seroconvert compared to individuals who did not report this behavior. Men who have sex with men who reported that their last sex partner was the same race/ethnicity had a higher hazard of seroconversion than MSM who reported a different race/ethnicity. More sexual partners in the last three months also led to a higher hazard of seroconversion. Intimate partner violence also had a marginally significant association with seroconversion. Lastly, individuals who reported using methamphetamine and/or inhaled nitrates in the past year were more likely to seroconvert. The only variables in the final model not significantly associated with HIV seroconversion were age group, age difference of the last sex sexual partner, and ecstasy use.
Construction of the Risk Score
After risk score creation, a table was created showing the total number of MSM included, number of HIV-positive MSM, HIV-negative MSM who had scores greater than or equal to each HIV risk score cutpoint, and the number of HIV infections per 100 person-years for each risk score (Table 6). Approximately 51% of all MSM had a risk score greater than or equal to 5.
If all individuals in our population who had a risk score greater than or equal to 5 (51%) had been given PrEP, 75% of HIV infections would be averted during follow-up, assuming adequate regimen adherence and near complete effectiveness. In comparison, the CDC criteria would recommend PrEP for 69% of all MSM in this data set averting 86% of all infections provided positives were instead given PrEP and maintained appropriate adherence. Therefore, the additional variables proposed allow for more targeted PrEP use than current CDC criteria. Physicians, their patients, and other interested individuals can obtain their own personalized risk score by visiting www.IsPrEPforMe.org.
Comparison to Other Models
Harrell's C-index on the 100 hold-out samples was 0.6 showing adequate model predictive accuracy. We also compared our risk score model with CDC criteria and risk score models by Menza et al. and Smith et al. (Table 7). Our model had an AIC of 6094, whereas the model that generated the CDC criteria had an AIC of 6162.2 The BIC for our model (6160) also demonstrated a better fit than the CDC model (6170). The additional predictors included in the present study provide a significantly better fit to the data compared with the original CDC model (P < 0.0001).
This study found that HIV infection at follow-up was significantly associated with baseline race/ethnicity; self-reported history of chlamydia, gonorrhea, and/or syphilis; condom use during last receptive anal sex; same race/ethnicity of the last sex partner; number of sex partners in the last three months; IPV; methamphetamine use and nitrates use. We created an empirically based sexual risk score based on study measures. Our calculations indicated that treating 51% of the population with PrEP had the potential to prevent up to 75% of the HIV infections during follow-up at the chosen cutpoint of 5, provided that participants both initiate PrEP and adhere to the recommended regimen during the entire period that their risk remains significant. A previous mathematical model suggested that treating 50% of a cohort with PrEP would avert 29% of HIV infections over the next 20 years, given medication adherence levels seen in the preexposure prophylaxis initiative study.19,20
The better model fit of the current study compared to current CDC criteria shows that predictions of who will seroconvert are more reliable when partner-level variables are considered in addition to STI history and condom use variables. Further, when a client is unsure of their PrEP candidacy, this additional information can improve the guidance given about HIV risk and advisability of PrEP initiation.
Given our findings, we recommend PrEP for MSM with a risk score greater than or equal to 5 (equivalent in our sample to approximately 2 HIV infections per 100 person-years). If they have a risk score of 8 or higher, we strongly recommend PrEP (equivalent in our sample to approximately 3 infections per 100 person-years). For individuals whose risk score is less than 5 but who request PrEP, we believe the provider should consider it in light of their patient's overall concerns.
Our study has a number of limitations in both the data set and analysis. First, although AIC and BIC were compared between the Menza et al. and Smith et al. studies and the CDC criteria, some variables in the previous studies were not identical to those in the current analysis. Further, a serious limitation with the comparison is that our model was developed on our data, whereas the other models were developed on other data sets, and this provides a bias of unknown size in favor of our model. Second, 6.1% of the population was omitted from the final multivariable model due to missing values one or more variables. Third, recall bias is likely present because certain experiences may be easier to remember than others.
Fourth, there are numerous variables for which we had no data but which probably have a quantifiable relationship with HIV incidence. For example, sexual compulsivity and childhood sexual abuse have recently been shown to have a relationship with HIV acquisition.21–26 In addition, population-level variables in the region where one meets sexual partners, including community viral load and STI prevalence, also impact HIV incidence and should be considered. In addition, certain variables that were measured may be proxies for other variables that were not available for this analysis. For example, we found that racial/ethnic concordance with the last sex partner was associated with higher HIV risk. This variable may actually be a proxy for breadth of sexual network and more nuanced analyses are needed to explore this relationship.
Lastly, PrEP-specific factors such as willingness to take a daily medication, medication adherence, and concern over side effects should also be considered. Data on these variables were not collected during the analysis period, but future studies should evaluate how these psychosocial, population-level, and PrEP-specific predictors can be used to further hone calculations of HIV acquisition risk and subsequently PrEP candidacy.
Lastly, the development of this score was geographically limited to the Los Angeles area. Furthermore, both the current study and the original Menza study used samples collected from the Western Coast of the United States. Validation of these measures are needed in areas where the HIV epidemic is concentrated the most such as cities in the Southern United States.
Despite these limitations, our study has a number of strengths. Previous HIV risk models have had a dearth of racial/ethnic minority subjects which represent the majority of new HIV infections.3,4 In contrast, 52% of our study’s sample identified as a racial/ethnic minority. Furthermore, our analyses were informed by a well-tested theory for HIV acquisition in MSM. Syndemics theory, validated in numerous populations of MSM, introduced a psychosocial component which likely impacts HIV incidence and should be explored, especially for racial/ethnic minority MSM. The fit of our model was significantly better relative to CDC guidelines. We have shown that partner-level variables must be considered alongside STI history and substance use variables to build a truly comprehensive risk score. Another strength is that potential PrEP users can use our interactive website to predict their HIV risk and thus PrEP candidacy (www.IsPrEPforMe.org) so that they can be prepared before discussing with their medical provider. Lastly, and most importantly, we have proposed more defined PrEP criteria for MSM when compared to current CDC criteria. Although our model does not elucidate every factor that should be considered in determining PrEP initiation, it does represent an improvement over existing CDC guidelines and it points the way for inclusion of ever more relevant variables to permit increasingly more informed decisions about PrEP use.
Our proposed HIV risk algorithm provides a systematic tool for both providers and patients in the biomedical prevention era. Providers will be able to make a more personalized recommendation for each client. Clients will be better informed about what actions and circumstances specifically lead to their increased HIV risk. This, in turn can be empowering and engender a sense of personal agency for each client so they can better decide to initiate PrEP, reduce sexual risk, or make a plan that incorporates both PrEP initiation and sexual risk reduction.
1. Centers for Disease Control and Prevention. HIV among Gay and Bisexual Men. Available at: http://www.cdc.gov/hiv/risk/gender/msm/facts/index.html
. Accessed July 25, 2014.
2. US Public Health Service. Preexposure prophylaxis for the prevention of HIV infection in the United States - 2014 A clinical practice guideline. Available at: http://www.cdc.gov/hiv/pdf/prepguidelines2014.pdf
. Accessed September 30, 2014.
3. Menza TW, Hughes JP, Celum CL, et al. Prediction of HIV acquisition among men who have sex with men. Sex Transm Dis 2009; 36:547–555.
4. Smith DK, Pals SL, Herbst JH, et al. Development of a clinical screening index predictive of incident HIV infection among men who have sex with men in the United States. J Acquir Immune Defic Syndr 2012; 60:421–427.
5. Singer M. AIDS and the health crisis of the U.S. urban poor; the perspective of critical medical anthropology. Soc Sci Med 1994; 39:931–948.
6. Mustanski B, Garofalo R, Herrick A, et al. Psychosocial health problems increase risk for HIV among urban young men who have sex with men: Preliminary evidence of a syndemic in need of attention. Ann Behav Med 2007; 34:37–45.
7. Ayala G, Bingham T, Kim J, et al. Modeling the impact of social discrimination and financial hardship on the sexual risk of HIV among Latino and Black men who have sex with men. Am J Public Health 2012; 102:S242–S249.
8. Jie W, Ciyong L, Xueqing D, et al. A syndemic of psychosocial problems places the MSM (men who have sex with men) population at greater risk of HIV infection. PLoS One 2012; 7:e32312.
9. Hall HI, Byers RH, Ling Q, et al. Racial/ethnic and age disparities in hiv prevalence and disease progression among men who have sex with men in the United States. Am J Public Health 2007; 97:1060–1066.
10. Centers for Disease Control and Prevention. HIV among African American gay and bisexual men. Available at: http://www.cdc.gov/hiv/group/msm/bmsm.html
. Accessed December 28, 2015.
11. Centers for Disease Control and Prevention. HIV among Hispanics/Latinos. Available at: http://www.cdc.gov/hiv/group/racialethnic/hispaniclatinos/
. Accessed December 28, 2015.
12. Smirnov N. Table for estimating the goodness of fit of empirical distributions. Ann Math Stat 1948; 19:279–279.
13. Grønnesby JK, Borgan O. A method for checking regression models in survival analysis based on the risk score. Lifetime Data Anal 1996; 2:315–320.
14. Wilson PWF, D'Agostino RB, Levy D, et al. Prediction of coronary heart disease using risk factor categories. Circulation 1998; 97:1837–1847.
15. Wang TJ, Massaro JM, Levy D, et al. A risk score for predicting stroke or death in individuals with new-onset atrial fibrillation in the community: The Framingham Heart Study. JAMA 2003; 290:1049–1056.
16. O'Seaghdha CM, Lyass A, Massaro JM, et al. A risk score for chronic kidney disease in the general population. Am J Med 2012; 125:270–277.
17. Harrell FE, Lee KL, Mark DB. Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996; 15:361–387.
18. Akaike H. New look at statistical-model identification. IEEE T Automat Contr 1974 AC19(6):716–723.
19. Grant RM, Lama JR, Anderson PL, et al. Preexposure chemoprophylaxis for HIV prevention in men who have sex with men. N Engl J Med 2010; 363:2587–2599.
20. Juusola JL, Brandeau ML, Owens DK, et al. The cost-effectiveness of preexposure prophylaxis for HIV prevention in the United States in men who have sex with men. Ann Intern Med 2012; 156:541–550.
21. Mimiaga MJ, Noonan E, Donnell D, et al. Childhood sexual abuse is highly associated with HIV risk-taking behavior and infection among MSM in the EXPLORE Study. J Acquir Immune Defic Syndr 2009; 51:340–348.
22. Sweet T, Welles SL. Associations of sexual identity or same-sex behaviors with history of childhood sexual abuse and HIV/STI risk in the United States. J Acquir Immune Defic Syndr 2012; 59:400–408.
23. Phillips G 2nd, Magnus M, Kuo I, et al. Childhood sexual abuse and HIV-related risks among men who have sex with men in Washington, DC. Arch Sex Behav 2014; 43:771–778.
24. Grov C, Parsons JT, Bimbi DS. Sexual compulsivity and sexual risk in gay and bisexual men. Arch Sex Behav 2010; 39:940–949.
25. Parsons JT, Grov C, Golub SA. Sexual compulsivity, co-occurring psychosocial health problems, and HIV risk among gay and bisexual men: Further evidence of a syndemic. Am J Public Health 2012; 102:156–162.
26. Woolf-King SE, Rice TM, Truong H-HM, et al. Substance use and HIV risk behavior among men who have sex with men: The role of sexual compulsivity. J Urban Health 2013; 90:948–952.