A Syndemic of Psychosocial Health Disparities and Associations With Risk for Attempting Suicide Among Young Sexual Minority Men

Abstract

Objectives. We examined a syndemic of psychosocial health issues among young men who have sex with men (MSM), with men and women (MSMW), and with women (MSW). We examined hypothesized drivers of syndemic production and effects on suicide attempts.

Methods. Using a pooled data set of 2005 and 2007 Youth Risk Behavior Surveys from 11 jurisdictions, we used structural equation modeling to model a latent syndemic factor of depression symptoms, substance use, risky sex, and intimate partner violence. Multigroup models examined relations between victimization and bullying experiences, syndemic health issues, and serious suicide attempts.

Results. We found experiences of victimization to increase syndemic burden among all male youths, especially MSMW and MSM compared with MSW (variance explained = 44%, 38%, and 10%, respectively). The syndemic factor was shown to increase the odds of reporting a serious suicide attempt, particularly for MSM (odds ratio [OR] = 5.75; 95% confidence interval [CI] = 1.36, 24.39; P < .001) and MSMW (OR = 5.08; 95% CI = 2.14, 12.28; P < .001) compared with MSW (OR = 3.47; 95% CI = 2.50, 4.83; P < .001).

Conclusions. Interventions addressing multiple psychosocial health outcomes should be developed and tested to better meet the needs of young MSM and MSMW.

Substantial evidence has been accumulating over the past several decades to suggest that men who have sex with men (MSM) experience substantial disparities in many facets of health.¹ An important characteristic of these disparities is that they tend to arise early in the life course. For example, meta-analyses have shown that by adolescence and young adulthood MSM experience significantly higher rates of depression,² substance use,³ HIV sexual risk behaviors,⁴ and suicidality² than do their heterosexual peers. Little research has been conducted on the health of young men who have sex with men and women (MSMW). The few empirical studies separating adolescent MSM from adolescent MSMW have shown that adolescent MSMW report significantly higher rates of substance use, including tobacco use, than do their heterosexual³ and MSM^5–9 counterparts. Additionally, MSMW have been found to be more likely to report HIV sexual risk behaviors than are their heterosexual peers.⁹

One theoretical explanation for these disparities is syndemics.^10–12 Syndemic theory posits that as individuals are confronted with adversity across the life course, particularly in the forms of social marginalization and victimization, they develop psychosocial health problems such as low self-image, depression, and substance use.¹³ These conditions tend to co-occur, which has a snowballing effect on overall health. In fact, several studies have demonstrated that as the number of psychosocial conditions within an individual increases, so does their risk of major negative health outcomes.^10–12,14 It follows from this argument that young MSM, who experience far greater levels of adversity than do heterosexual youths,^15–18 would also experience greater rates of syndemics and higher rates of the resulting negative health outcomes.

We examined syndemic production in a population-based sample of young men and its association with serious suicide attempts. By looking at syndemics in a large enough sample that could be broken into MSM, MSMW, and men who have sex with women (MSW), we sought to determine whether syndemics are a general human phenomenon or whether they exist and are associated with negative health outcomes only for certain socially marginalized populations.^10,11,19 We also examined the structure of the syndemic construct by testing for measurement invariance using multigroup confirmatory factor analysis (CFA),²⁰ which, essentially, asks whether the meaning of the syndemic construct is the same across groups. Next we examined the relation between experienced adversity and syndemic production to see whether this association exists for all groups of young men, and if so, to what degree. Finally, we tested the relationship between syndemics and a serious adverse health outcome—making a life-threatening suicide attempt—and whether the negative effects were the same across groups.

METHODS

We gathered data for this study by pooling Youth Risk Behavior Surveillance System surveys collected during 2005 and 2007 from several US jurisdictions. The general approach to pooling and analyzing the data, as well as detailed information about the sexual orientation items and characteristics of the jurisdictions, are described elsewhere in this issue.²¹ We analyzed data from the 11 jurisdictions that measured gender of sexual partners, including Boston, Massachusetts; Chicago, Illinois; Delaware; Maine; Massachusetts; New York City, New York; San Diego, California; Vermont; Rhode Island; Wisconsin; and Milwaukee, Wisconsin.

Measures

Participants were asked to report during their lifetime if they had sexual contact (or, in some jurisdictions, “sexual intercourse”) with female, male, or male and female partners, or if they had no sexual partners. Among male respondents, we recoded this variable as MSM, MSMW, and MSW.

Syndemic indicators.

We used 6 variables as syndemic indicators: depressive symptoms, binge drinking, regular marijuana use, cocaine use, risky sexual behavior, and intimate partner violence. We dichotomized each variable so that a score of 1 represented a clear problematic level of the syndemic indicator. We used this approach to be consistent with syndemic theory, which focuses on the unhealthy end of the distribution of outcomes, rather than normal variation. We determined cutoffs for each variable by examining the distribution of the responses, prior literature, and clinical significance. We treated binge drinking, regular marijuana use, and cocaine use as separate syndemic indicators, rather than creating a polydrug use indicator as has been done in some prior syndemic studies,^11,12 to be consistent with prior syndemic studies among young MSM^10,22 and because these substances have unique pharmacological effects, patterns and contexts of use, and relationships with health outcomes.

Binge drinking, which was defined as “5 or more drinks of alcohol in a row, that is, in a couple of hours,” was measured in number of days in the prior 30 days. We coded individuals as positive for regular binge drinking if they reported 3 or more days of binge drinking. We assessed other substance use variables as the number of times the substance was used in the prior 30 days. We coded regular marijuana use as positive if respondents indicated using marijuana 20 or more times in the past 30 days. We coded cocaine use as positive if participants reported using cocaine 3 or more times in the past 30 days.

We measured risky sexual behavior with 2 items. The first was a measure of the number of sexual partners, “During the past 3 months, with how many people did you have sexual intercourse?” with response options including “I have never had sexual intercourse,” “I have had sexual intercourse, but not during the past 3 months,” “1 person,” “2 people,” up to “6 or more people.” Additionally, we assessed condom use during the last sexual event, “The last time you had sexual intercourse, did you or your partner use a condom?” with the response options including “I have never had sex,” “yes,” and “no.” We constructed a dichotomized variable to assess sexual risk, with no condom use during the last sexual event and 2 or more sexual partners during the past 3 months indicating risky sexual behavior.

We assessed intimate partner violence with an item that asked, “During the past 12 months did your boyfriend or girlfriend ever hit, slap, or physically hurt you on purpose?” with “yes” or “no” response options. We assessed depressive symptomatology using an item that asked, “During the past 12 months, did you ever feel so sad or hopeless almost every day for 2 weeks or more in a row that you stopped doing some usual activities?” Response options were “yes” or “no.”

Hypothesized syndemic drivers.

We selected correlates of the health issues on the basis of Stall’s theory of syndemic production among MSM.¹³ Variables included being attacked with a weapon, having personal property stolen or deliberately damaged, and being physically forced to have sex. Respondents were asked, “During the past 12 months, how many times has someone threatened or injured you with a weapon such as a gun, knife, or club on school property?” and “During the past 12 months, how many times has someone stolen or deliberately damaged your property such as your car, clothing, or books on school property?” We then dichotomized these 2 variables with a positive indication representing occurrence of these events 1 or more times. To assess forced sex, respondents were asked to indicate “yes” or “no” to the following question: “Have you ever been physically forced to have sexual intercourse when you did not want to?”

Serious suicide attempt.

Serious suicide attempt was assessed by asking participants, “If you attempted suicide during the past 12 months, did any attempt result in an injury, poisoning, or overdose that had to be treated by a doctor or nurse?” Respondents were asked to indicate “I did not attempt suicide in the past 12 months,” “yes,” or “no.”

Statistical Analyses

We calculated prevalences and relative risks (RRs) among observed variables for MSM, MSMW, and MSW with SPSS version 20 (SPSS Inc, Chicago, IL). We used Mplus version 7.0 (Muthén & Muthén, Los Angeles, CA) to calculate tetrachoric correlations and run a series of multigroup structural equation models (SEMs) to examine the hypothesized relations between victimization and bullying experiences, syndemic health issues, and serious suicide attempts. Structural equation modeling allows the specification and estimation of relationships between observed and latent variables²³ and the evaluation of measurement and structural invariance through an examination of model parameters and fit indices.²⁴ This approach is particularly well suited to the analysis of a syndemic model because syndemic theory posits an underlying construct that represents the association among a number of health issues,^13,25,26 which can be operationalized as a latent variable.²⁷ A latent variable modeling approach has the advantage of reducing the dimensionality of the data by capturing variance shared across the observed indicators, of eliminating measurement error, and of using a more sophisticated approach for weighting the contribution of health issues to the underlying construct rather than assuming equal contributions.²³ This approach is commonly implemented in the study of health issues that are believed to have an underlying association, such as externalizing disorders²⁸ or problem behaviors among adolescents.²⁹ The advantage of the multigroup modeling approach is that it allows us to explicitly test differences in the structure of the syndemic and associations with independent and dependent variables between MSM, MSMW, and MSW.

We tested the overall configuration of the syndemic latent factor with 6 dichotomous indicators using multigroup CFA with weighted least squares mean and variance-adjusted estimation. This approach to latent variable modeling with dichotomous data is akin to an item response theory (IRT) approach.³⁰ We assessed model fit for the CFAs using the root mean square error of approximation (RMSEA),³¹ comparative fit index (CFI),³² and Tucker–Lewis index (TLI).³³ We tested factorial invariance across groups and nested models with a mean and variance-adjusted χ² statistic using the Mplus DIFFTEST command.³⁴

We entered the victimization and suicide variables into the model to comprise the structural portion of the model, and we used a multigroup approach to compare differences between MSM, MSMW, and MSW. We used maximum likelihood estimation with robust SEs to perform logistic regression in the multigroup SEM analyses. To include the maximum likelihood estimation with a robust SEs estimator in a multigroup SEM model with dichotomous variables in Mplus, we used the TYPE = MIXTURE and KNOWNCLASS commands. We determined model selection by comparisons between the Akaike information criterion³⁵ and the Bayesian information criterion,³⁶ with lower values indicating better fit, and the calculated scaled χ² difference test.³⁷

All analyses accounted for the complex sampling design of the Youth Risk Behavior Surveillance System by including sample weights, primary sampling units, and stratum variables. We calculated percentages and RR coefficients using the SPSS complex samples module. In Mplus, we included weight, cluster, and stratification commands to calculate correlations and for SEM. We accommodated missing data with pairwise deletion in SPSS and when using the weighted least squares mean and variance-adjusted estimation estimator in the Mplus measurement models. For the SEM structural models analyzed in Mplus with the MLR estimator, a maximum likelihood estimation approach under the missing at random assumption was utilized.

RESULTS

The sample consisted of 16 977 male respondents aged 13 years or older who answered 1 or more questions indicating they were sexually active with male partners, female partners, or both. The sample ranged from aged 13 through aged 18 years (mean = 16.14). Of the male youths, 4.0% (n = 608) were MSM and 3.2% (n = 577) were MSMW. The sample was racially and ethnically diverse, with 51.4% identifying as White, 22.3% as Black or African American, 14.5% as Hispanic or Latino, 7.5% as multiracial or other, 3.0% as Asian, and 1.3% as American Indian.

Table 1 presents the prevalence and RRs for all observed study variables, treating MSW as the reference group. Notable in the table is the extent to which MSMW reported higher rates of all variables relative to MSW, with RRs ranging from 1.71 to 8.20 and most at least twice as high. The results were different for MSM, who had significantly higher RRs than did MSW for half of the syndemic indicators (felt sad, cocaine use, intimate partner violence), all the hypothesized syndemic drivers, and the primary outcome of a serious suicide attempt. The prevalence of all syndemic indicators was higher among MSMW than among MSM.

TABLE 1—

Weighted Prevalence and Relative Risks of Observed Variables for MSW, MSMW, and MSM From 11 Jurisdictions That Collected Gender of Sexual Partners: Youth Risk Behavior Surveillance System, United States, 2005 and 2007

	%			RR (95% CI)
Variable	MSW	MSMW	MSM	MSW and MSMW	MSW and MSM
Felt sad	20.7	44.2	34.7	2.13* (1.78, 2.56)	1.67* (1.37, 2.03)
Binge drinking	15.3	27.6	16.3	1.81* (1.43, 2.29)	1.07 (0.76, 1.49)
Marijuana use	11.0	24.8	6.7	2.26* (1.66, 3.08)	0.61 (0.36, 1.05)
Cocaine use	2.4	19.4	8.0	8.20* (5.06, 13.29)	3.39* (1.80, 6.38)
Risky sex	4.9	27.3	4.9	5.56* (3.86, 8.00)	1.00 (0.47, 2.13)
Intimate partner violence	14.0	30.0	21.0	2.14* (1.69, 2.70)	1.50* (1.14, 1.98)
Threatened with weapon	9.8	28.5	14.2	2.91* (2.29, 3.71)	1.46* (1.04, 2.04)
Had something stolen	26.6	45.6	34.4	1.71* (1.48, 1.98)	1.29* (1.06, 1.57)
Forced to have sex	6.7	33.1	19.9	4.95* (4.05, 6.05)	2.97* (2.21, 3.99)
Suicide attempt	2.2	15.1	6.9	6.74* (4.34, 10.46)	3.09* (1.71, 5.58)

Note. CI = confidence interval; MSM = men who have sex with men; MSMW = men who have sex with men and women; MSW = men who have sex with women; RR = relative risk. The 11 jurisdictions were Boston, MA; Chicago, IL; Delaware; Maine; Massachusetts; New York City, NY; San Diego, CA; Vermont; Rhode Island; Wisconsin; and Milwaukee, WI.

*P < .05.

The bivariate tetrachoric correlation matrix is provided in Table 2. Variable indicators for the latent syndemic factor were all positively correlated with small to moderate correlations (with the exception of binge drinking and feeling sad for MSM). It is of note that we found larger correlations among the latent variable indictors for the MSMW than the MSW. These results suggest that although the syndemic health items were interrelated across the 3 groups, they were more strongly associated with each other among MSMW than among MSW. The correlations among MSM were intermediate between MSW and MSMW for most variables.

TABLE 2—

Tetrachoric Correlations of Model Variables Among MSW, MSMW, and MSM From 11 Jurisdictions That Collected Gender of Sexual Partners: Youth Risk Behavior Surveillance System, United States, 2005 and 2007

Variable	1	2	3	4	5	6	7	8	9
MSW
Felt sad	…
Binge drinking	0.09*	…
Marijuana use	0.12*	0.57*	…
Cocaine use	0.23*	0.65*	0.71*	…
Risky sex	0.11*	0.42*	0.42*	0.60*	…
Intimate partner violence	0.19*	0.11*	0.17*	0.32*	0.27*	…
Threatened with weapon	0.33*	0.17*	0.15*	0.41*	0.26*	0.21*	…
Had something stolen	0.26*	0.10*	0.10*	0.27*	0.16*	0.16*	0.45*	…
Forced to have sex	0.21*	0.17*	0.26*	0.44*	0.39*	0.48*	0.31*	0.21*	…
Suicide attempt	0.36*	0.30*	0.32*	0.63*	0.33*	0.20*	0.38*	0.26*	0.40*
MSMW
Felt sad	…
Binge drinking	0.37*	…
Marijuana use	0.30*	0.83*	…
Cocaine use	0.10	0.71*	0.87*	…
Risky sex	0.16*	0.65*	0.77*	0.94*	…
Intimate partner violence	0.30*	0.61*	0.63*	0.68*	0.62*	…
Threatened with weapon	0.25*	0.41*	0.57*	0.66*	0.47*	0.48*	…
Had something stolen	0.33*	0.39*	0.29*	0.54*	0.24*	0.47*	0.67*	…
Forced to have sex	0.44*	0.64*	0.70*	0.60*	0.60*	0.85*	0.57*	0.48*	…
Suicide attempt	0.66*	0.57*	0.68*	0.81*	0.62*	0.66*	0.70*	0.55*	0.74*
MSM
Felt sad
Binge drinking	−0.17*
Marijuana use	0.21*	0.58*
Cocaine use	0.33*	0.49*	0.78*
Risky sex	0.12	0.46*	0.43*	0.52*
Intimate partner violence	0.26*	0.40*	0.46*	0.42*	0.16
Threatened with weapon	0.21*	0.40*	0.41*	0.67*	0.29*	0.66*
Had something stolen	0.09	0.44*	0.23*	0.63*	0.47*	0.14	0.51*
Forced to have sex	0.33*	0.31*	0.23*	0.41*	0.22	0.35*	0.43*	0.12
Suicide attempt	0.43*	0.48*	0.61*	0.85*	0.65*	0.73*	0.76*	0.41*	0.74*

Note. MSM = men who have sex with men; MSMW = men who have sex with men and women; MSW = men who have sex with women. We produced correlations in Mplus with a model that accounted for the complex sampling design. The 11 jurisdictions were Boston, MA; Chicago, IL; Delaware; Maine; Massachusetts; New York City, NY; San Diego, CA; Vermont; Rhode Island; Wisconsin; and Milwaukee, WI.

*P < .05.

To evaluate the extent to which the variables formed a syndemic latent factor, we analyzed CFA models for the full sample of sexually active male youths. For identification purposes, we set the variance of the latent factor to 1 for all analyses.³⁸ Consistent with syndemic theory,^13,25,26 a 1-factor CFA model was fit to the data with all 6 health items loading on the syndemic factor (Figure 1, paths a–f). Fit indices met specification criteria as recommended by Hu and Bentler³⁹ (RMSEA = 0.02; CFI = 0.97; TLI = 0.96), signifying that the 1-factor model fit the data well.

FIGURE 1—

Multigroup structural equation model of victimization experiences, syndemic health disparities, and serious suicide attempts: Youth Risk Behavior Surveillance System, United States, 2005 and 2007.

Note. AIC = Akaike information criterion; BIC = Bayesian information criterion; IPV = intimate partner violence. Structural model 1: configural invariance (paths a–j free across groups): AIC = 55986.18; BIC = 56384.96; log-likelihood = −27940.09. Structural model 2: strong invariance (paths g–j restricted across groups): AIC = 56059.33; BIC = 56382.87; log-likelihood = −27986.64. Model 1 vs 2: scaled χ² difference test (Δχ² = 35.18; df = 10; P < .001).

Because of the group differences in prevalences and interindicator correlations, we conducted a series of multigroup CFAs to test for invariance of the measurement model. The first model tested configural invariance in which factor loadings and item thresholds were freely estimated across groups, with all item scale factors fixed to 1 and factor means fixed to zero for identification purposes.⁴⁰ Fit indices for the configural invariance model indicated that the model fit the data well (RMSEA = 0.02; CFI = 0.98; TLI = 0.96). Achieving this level of invariance demonstrated that the basic factor structure fit the data in each group.

We then compared the configural invariance model with a strong invariance model that constrained the factor loadings and item thresholds parameters to be equal across groups. For identification purposes, we fixed the scale factors to 1 in 1 group and to free in the others and the factor means to zero in 1 group and to free in the others.⁴⁰ The strong invariance model was a significantly worse fit to the data than was the configural invariance model (Δχ² = 64.07; Δdf = 10; P < .001). This finding suggests that although the same items load on the syndemic factor for all groups, the items’ factor loadings significantly differed and therefore the meaning of the latent syndemic factor also differed between the groups.

The next set of analyses focused on the structural relation between experiences of victimization and bullying, the syndemic factor, and a history of serious suicide attempts. We conducted a series of multigroup SEM analyses to test for configural and strong invariance of the structural model between MSM, MSMW, and MSW (Figure 1). The configural invariance model allowed both the measurement (paths a–f) and structural (paths g–j) parameters to be freely estimated across groups, whereas the strong invariance model restricted the structural parameter estimates (paths g–j) to be equal across groups. Fit indices for the 2 models are provided in Figure 1. The baseline model was found to be the besting fitting model because of the lower Akaike information criterion and Bayesian information criterion values and significant scaled χ² difference test. This finding indicates that although the basic model structure fit the data well, the structural path coefficients differed between groups, and therefore the effects of the hypothesized drivers on the syndemic construct and the effect of the syndemic on the suicide attempt outcome differed between groups. Parameter estimates and 95% confidence intervals (CIs) for the final (configural) model are presented in Table 3. Note that consistent with configural invariance, the size of the factor loadings differed between the groups. For example, marijuana use was more central to the syndemic factor among MSW and MSMW than among MSM. Only 2 syndemic indicators had nonsignificant loadings among all the groups; for MSM the felt sad and cocaine use variables were nonsignificant. The felt sad variable had small R² values for all groups, so this indicator is less central to the syndemic across groups. The cocaine use variable actually had a large R² across all groups, but the relatively low base rate and smaller size of the MSM sample appears to have produced wide CIs and a nonsignificant loading. The significant differences between MSMW and MSW in the measurement of the syndemic are exemplified by the greater proportion of variance explained in most health issues by the latent syndemic among the MSMW (e.g., 63% among MSMW and 44% among MSW for binge drinking). The least variance was explained in the syndemic indicators among MSM. Because of lack of evidence for strong invariance in the measurement of the syndemic constructs across the groups, it is important to interpret analyses of hypothesized drivers and consequences in light of the different configuration of the syndemic variable across groups.

TABLE 3—

Unstandardized Model Parameter Estimates for MSW, MSMW, and MSM From 11 Jurisdictions That Measured Gender of Sexual Partners: Youth Risk Behavior Surveillance System, United States, 2005 and 2007

		MSW			MSMW			MSM
Parameters	Path	ba (95% CI)	SE	R2	ba (95% CI)	SE	R2	ba (95% CI)	SE	R2
Felt sad	a	0.48 (0.38, 0.60)	0.06	0.07	0.52 (0.18, 0.85)	0.17	0.13	0.23c (−0.33, 0.80)	0.29	0.03
Binge drinking	b	1.51 (1.21, 1.82)	0.16	0.44	1.76 (0.65, 2.87)	0.57	0.63	0.96 (0.12, 1.81)	0.43	0.31
Marijuana use	c	1.80 (1.36, 2.25)	0.23	0.52	2.58 (0.85, 4.31)	0.88	0.78	0.96 (0.09, 1.82)	0.44	0.31
Cocaine use	d	4.01 (2.16, 5.87)	0.95	0.85	3.41 (0.29, 6.20)	1.43	0.86	2.29c (−0.30, 4.87)	1.32	0.72
Risky sex	e	1.35 (1.09, 1.62)	0.14	0.38	1.48 (0.29, 2.66)	0.60	0.54	0.65 (0.24, 1.06)	0.21	0.17
Intimate partner violence	f	0.68 (0.53, 0.84)	0.08	0.14	2.01 (1.15, 2.88)	0.44	0.69	0.84 (0.10, 1.59)	0.28	0.26
Threatened with a weapon	g	0.58 (0.43, 0.74)	0.08	…	0.58 (0.03, 1.13)	0.28	…	1.25 (0.48, 2.02)	0.29	…
Had something stolen	h	0.23 (0.11, 0.35)	0.06	…	0.22c (−0.20, 0.64)	0.21	…	0.82 (0.15, 1.48)	0.34	…
Forced to have sex	i	0.90 (0.69, 1.11)	0.11	…	1.44 (0.87, 2.01)	0.29	…	0.90c (−0.18, 1.99)	0.55	…
Suicide attempt	j	3.47b (2.50, 4.83)	0.17	0.34	5.08b (2.14, 12.28)	0.45	0.59	5.75b (1.36, 24.39)	0.74	0.60
Syndemic latent factor	…	…	…	0.10	…	…	0.44	…	…	0.38

Note. CI = confidence interval; MSM = men who have sex with men; MSMW = men who have sex with men and women; MSW = men who have sex with women. We performed structural equation modeling in Mplus with a model that accounted for the complex sampling design. Unless otherwise indicated, parameter estimates are significant at P < .05. The 11 jurisdictions were Boston, MA; Chicago, IL; Delaware; Maine; Massachusetts; New York City, NY; San Diego, CA; Vermont; Rhode Island; Wisconsin; and Milwaukee, WI.

^aUnless otherwise indicated, estimates are presented as unstandardized b coefficients.

^bOdds ratio.

^cNonsignificant estimate.

As hypothesized, we found experiences of victimization and bullying to increase the syndemic burden experienced by all youths. However, these independent variables were significantly more influential for MSM and MSMW than for MSW, with more than 3 times the variance in the syndemic factor explained among MSM (38%) and MSMW (44%) than among MSW (10%). In addition, the syndemic factor was shown to significantly increase the odds of having a serious suicide attempt much more among MSM (odds ratio [OR] = 5.75; CI = 1.36, 24.39; P < .001) and MSMW (OR = 5.08; CI = 2.14, 12.28; P < .001) than among MSW (OR = 3.47; CI = 2.50, 4.83; P < .001). Overall, the final model explained 60% of the variance in suicide attempts for MSM, 59% for MSMW, and 34% for MSW.

DISCUSSION

We have confirmed earlier work that suggests that sexual minorities (MSM and MSMW) experience serious health disparities in adolescence and that these psychosocial health problems tend to co-occur and act to raise risk levels for poor health outcomes—in this case suicide attempts requiring medical attention. We have extended previous research to show that psychosocial health outcomes also co-occur among the majority group (MSW), but the measurement of the syndemic construct is not invariant across groups and therefore the meaning of the construct also differs. Furthermore, we found that higher levels of these syndemic components increased suicide attempts among all youths, but most strongly among the MSM and MSMW. This is somewhat surprising because a core risk factor for suicide is feelings of depression,⁴¹ which has the lowest association with the syndemic construct among the MSM. Nevertheless, the syndemic factor among the MSM and MSMW explained a very large proportion of the variance in making a serious suicide attempt. This pattern of results suggests that syndemics are not specific to nonheterosexual populations and are in fact a general human phenomenon, but their composition differs and the consequences are felt most deeply by those in the minority.

Limitations

There are several limitations that need to be considered when interpreting these results. First, we coded individuals as MSM or MSMW on the basis of their lifetime sexual contact. It is possible that many of the sexually active young men who were struggling to accept their homosexuality, but who have not yet been sexually active with male partners or do not wish to acknowledge same-sex sexual behavior, were incorrectly coded as MSW, thereby attenuating the differences between the MSM and MSMW group and the MSW group.⁴²

Additionally, sexual activity was an inclusion criterion, which may have affected our findings. Also, the 11 jurisdictions that we included in this pooled analysis of Youth Risk Behavior Surveillance System data were those that opted to include a measure of sexual behavior with same-sex partners. The jurisdictions that ask this information are most likely areas where sexual minorities face less day-to-day marginalization and are therefore faring much better than are those in less-tolerant climates.^43–45 These results may therefore not be generalizable across geographical locales and may, in fact, provide a conservative estimate of the extent and health outcomes of syndemics among MSM and MSMW youths.

These analyses were also cross-sectional, and therefore conclusions cannot be drawn about causality or prediction of outcomes. A limitation inherent in most secondary analysis of surveillance data is that we could not select measures specific to our research question, and our variables were mostly derived from single items rather than scales. These analyses also need to be replicated in young women and transgender youth populations before they can be generalized to other sexual minority populations.

Conclusions

This study extends the literature in several important ways. First, there have been multiple studies that have replicated the finding that syndemic clusters exist in sexual minorities and that these clusters arise early in life. However, we have demonstrated that syndemics are not unique to sexual minorities but also exist in a different form among heterosexual youths. A novel aspect of our study was breaking apart MSM from MSMW, and our results show that these syndemic clusters co-occur with particular strength among MSMW.

Second, the majority of studies that have looked into syndemic production thus far have focused primarily on the effects of these clusters on HIV or sexual health–related outcomes.^{10–12,14,19,26} We have shown that syndemics also account for a significant amount of variance in serious suicide attempts among all groups, but again, syndemics appear to have a much larger impact on suicide attempts among MSM and MSMW youths than among MSW youths. One interesting finding from this study is that the prevalence rates of the victimization and syndemic variables were higher among the MSMW than among the MSM. Although this trend has been demonstrated previously among sexual minority women,^46–48 it is an understudied phenomenon among young men.

Finally, we found that much of the syndemic production in this sample is related to victimization (being threatened with a weapon, having property stolen, or experiencing forced sex). Although this relationship between victimization and syndemics has previously been demonstrated among adults,⁴⁹ this is the first study to our knowledge to look into this relationship among youths. Related to this finding we once again see the pattern we have described: the relationship between victimization and syndemics holds true for all groups; however, sexual minorities are significantly more likely to experience victimization—and more than 3 times more likely to experience the most severe form of victimization, forced sex. These findings together suggest that efforts to address health disparities vulnerable youths face would benefit by incorporating syndemic theory as a basis for intervention design.

We now have multiple studies^10–12,49 to demonstrate that (1) experienced victimization accounts for a significant portion of syndemic development, (2) syndemics account for significant variance in health outcomes, and (3) sexual minorities experience notable disparities in every step of this process: higher rates of victimization and higher rates of syndemics with greater impact on health outcomes. For example, in this analysis, MSM and MSMW reported much higher rates of serious suicide attempts than did MSW (6.2% and 17.9%, respectively, vs 2.5%).

Considering that these general findings have been replicated across numerous samples, the added value of further descriptive studies showing that sexual minority men are at increased risk for poor health outcomes may be limited. It is now time to turn our efforts to identifying mechanisms and developing and testing interventions that are designed to disrupt syndemic formation among vulnerable youths, particularly MSM and MSMW. To do this effectively we will need to incorporate structural factors that drive syndemic production into these interventions rather than focusing solely on the individual.⁵⁰ Interventions that address the context of adolescence, particularly those that focus on the school environment—where much of the victimization experienced by youths occurs⁵¹—are likely to yield larger effects than are those that intervene only with the persons being victimized.

Furthermore, interventions need to be specifically tailored with a special focus on racial/ethnic minority populations. Ample evidence exists to demonstrate heightened disparities in victimization, syndemics, and health outcomes among racial and ethnic minorities compared with their White peers.^52–54 Additionally, interventions that incorporate resiliencies into their theoretical underpinnings may show greater effects than do those that focus on deficits alone.^55–58 As an added benefit, interventions that focus on strengths and protective factors are likely to see greater participation and retention rates than do those that focus on negative attributes or unhealthy behaviors alone. Finally, interventions that focus on multiple psychosocial health conditions and test their effects on multiple outcomes may go a long way in raising the overall health of youths. Addressing this agenda could yield a practical tool to allow vulnerable and sexual minority youths to begin their adult years as thriving, healthy young men.