The methodological core of the study is what statisticians call multilevel latent class analysis, or MLCA. Traditional variable-centered approaches ask how single factors, such as partner age or substance use, relate to outcomes like condomless sex. Person-centered mixture models instead search for hidden subgroups within a population, defined by co-occurring patterns across many indicator variables simultaneously. The multilevel extension is crucial for sexual network data, because each participant, or ego, nominated multiple partners, or alters, creating a nested structure in which partnerships are clustered within individuals. Ignoring this nesting can distort estimates and obscure the fact that a single man may move fluidly between very different kinds of sexual encounters. The researchers modeled both levels at once: lower-level classes describing types of individual partnerships, and upper-level groups describing the overall composition of each man’s sexual network.

Data collection relied on an egocentric social network approach implemented through the Network Canvas platform, with participants recruited through a modified respondent-driven sampling strategy led by a community-based sexual health clinic. Each participant named the people with whom he had anal or vaginal sex in the past six months and reported detailed characteristics of each relationship: whether the partner was a main or casual partner, whether condoms were always used, which sexual positions involved condomless contact, whether the partner disclosed an HIV status, and how emotionally close the respondent felt to that person. Covariates included alcohol and other substance use when the dyad met to have sex, the respondent’s current PrEP use, country of nativity, and total sexual network size. Participants averaged 34.5 years of age, 88 percent were born outside the United States, and 57 percent reported currently using PrEP.

The analysis converged on a three-class, two-group solution. At the partnership level, three distinct sexual partner typologies emerged. The first, accounting for 37 percent of partnership contexts, was an elevated-risk context with a main sexual partner: seroconcordant dyads in which both partners disclosed being HIV-negative, marked by high emotional closeness, main partnership status, and a high probability of versatile condomless sex. The second class, 35 percent of contexts, was labeled selective and close casual partnerships, in which men reported condomless sex in only one position, such as insertive or receptive anal sex exclusively, alongside moderate emotional closeness. The third class, 27 percent, comprised low-exposure casual partnerships with little condomless sex, variable closeness, and a high probability that the partner’s HIV status was simply unknown. At the network level, two groups emerged: one dominated by casual partnerships, accounting for 54 percent of sexual partnerships, and one centered on a main sexual partner, accounting for 46 percent.

Perhaps the most striking finding concerns what these typologies reveal about trust. Consistent with the dyadic framework for HIV prevention developed by Karney and colleagues, the study found that risk behaviors clustered within emotionally close relationships rather than anonymous encounters. Condomless sex was most prevalent in main partnerships, where intimacy and trust may paradoxically substitute for barrier protection, and lowest in casual partnerships where serostatus was undisclosed. Partners in the low-exposure casual class frequently had unknown HIV seroconcordance, which may itself have prompted greater caution. Meanwhile, classes with higher perceived closeness involved partners who disclosed being HIV-negative, more versatile condomless practices, and more substance use when the dyad met for sex. This inversion of the intuitive risk hierarchy, in which the most trusted relationships carry the most condomless exposure, has been documented in prior research among Latino men and now receives a rigorous person-centered confirmation.

The multinomial logistic regression results sharpened the picture of who occupies which network context. Respondents currently using PrEP had 1.60 times greater odds of belonging to the group whose networks were composed primarily of main partnerships with elevated risk contexts, relative to the casual-network group. Latino men born outside the United States had 43 percent lower odds of having elevated-risk main-partner contexts compared to those born in the country. Network size mattered as well: every additional partner reduced the odds of belonging to the main-partnership group by 29 percent. Substance use when the dyad met to have sex was independently associated with roughly threefold higher odds of condomless sex in a random intercept model, and within the multilevel model, substance use sharply reduced the odds of membership in low-exposure casual classes relative to elevated-risk main-partner contexts.

These associations carry important caveats that the authors themselves emphasize. The data are cross-sectional, so the temporal direction of the PrEP finding remains unresolved: men may adopt PrEP because they are in higher-exposure partnerships, or their partnerships may take a particular shape because PrEP has reduced their perceived need for condoms. PrEP, when taken as directed, is extremely effective against HIV acquisition but does not prevent other sexually transmitted infections, and systematic reviews have documented elevated STI diagnosis rates among PrEP users, potentially reflecting both reduced condom use and increased testing frequency. The study also captured only the daily-pill form of PrEP, predating wider availability of long-acting injectable options, and the reporting window overlapped with the COVID-19 pandemic, when social distancing measures may have altered sexual network patterns in ways that differ from pre- and post-pandemic behavior.

Cultural and structural context adds further layers of complexity. Miami-Dade is a distinctive setting in which 69 percent of residents identify as Latino, more than half were born outside the United States, and 68 percent of adults speak Spanish. Approximately one-third of foreign-born participants had lived in the country for a year or less, a period that may be insufficient to establish close, serious partnerships and during which navigating new social and sexual environments can shape partner selection, communication, and trust. Prior research suggests that cultural values such as collectivism and familismo can support condom use, while acculturative processes, language barriers between partners, and hypermasculine norms may complicate safer-sex negotiation. The authors argue that future work should move beyond a simple US-born versus foreign-born dichotomy and apply syndemic theory or intersectional perspectives to capture how immigration experiences, stigma, substance use, and cultural values interact to shape vulnerability.

The practical implications point toward interventions designed at the level of the dyad and the network rather than the isolated individual. Couples-based programs can facilitate conversations about relationship agreements, the increasing openness of sexual partnerships, and the role of substances in sexual decision-making, while network interventions that engage opinion leaders can diffuse prevention messaging through the social structures the study has now characterized in detail. The researchers also propose expanding clinic screening tools to include brief egocentric sexual network modules, allowing providers to tailor recommendations, whether PrEP, doxycycline post-exposure prophylaxis, routine testing, or negotiated safety agreements, to the specific partnership typologies a patient actually experiences. By replacing blunt risk categories with empirically derived partnership profiles, the study offers a template for precision HIV and STI prevention that respects the fluid, context-dependent reality of Latino men’s sexual lives in one of America’s hardest-hit communities.

Subject of Research: Sexual partnership typologies and HIV/STI risk contexts among Latino men in Miami analyzed with multilevel latent class analysis of egocentric sexual network data

Article Title: Sexual Partner Typologies Among Latino Men Living in Miami, Florida: A Multilevel Latent Class Analysis

Article References: Despres, L., Rodriguez, E., Johnson, A. L., Bravo, A., Vilchez, L., Doblecki-Lewis, S., Fallon, S., & Kanamori, M. (2026). Sexual Partner Typologies Among Latino Men Living in Miami, Florida: A Multilevel Latent Class Analysis. Archives of Sexual Behavior. https://doi.org/10.1007/s10508-026-03441-1

Image Credits: AI Generated

DOI: 10.1007/s10508-026-03441-1

Keywords: HIV, sexually transmitted infections, Latino men, Miami-Dade County, multilevel latent class analysis, sexual networks, PrEP, condomless sex, substance use, dyadic prevention, immigration and acculturation, syndemic theory

The study tackles one of the most elusive challenges in substance use research: capturing the multifaceted, layered dimensions of how adolescents engage with various substances. Standard analyses often classify young individuals as simply users or non-users or focus solely on single substances, but this new framework transcends such limitations by recognizing simultaneous use patterns and the contextual factors influencing them. The MLCA method employed allows researchers to identify distinct subgroups, or latent classes, within the adolescent population that differ in their substance use behaviors, frequency, and combinations, thus depicting a more faithful portrait of reality.

Key to this analysis is the understanding that adolescent substance use is not monolithic but varies widely across multiple domains including type of substance, usage frequency, and associated behaviors. By integrating multilevel data, including individual traits and broader contextual influences such as socioeconomic status and regional differences, the research delves into the latent heterogeneity at both the individual and group levels. This dual lens enables the capture of subtle interplays between personal and environmental variables that shape substance use trajectories.

One of the particularly striking findings is the identification of several discrete classes of substance users among Canadian adolescents. These classes range from abstainers and experimental users to frequent multi-substance users exhibiting diverse risk profiles. For instance, some clusters are characterized by predominantly alcohol use, whereas others feature concurrent usage of alcohol, tobacco, and cannabis, revealing nuanced gradients of risk and consumption patterns.

Moreover, the study brings critical insight into how socio-demographic variables correlate with these latent classes. Differences based on factors such as age, gender, and geographical region were systematically integrated into the model, highlighting the social heterogeneity underpinning substance use behaviors. This comprehensive approach allows policymakers and prevention specialists to tailor interventions more precisely, focusing resources on subpopulations that are empirically identified as higher risk rather than relying on broad stigma or blanket policies.

The robustness of the multilevel latent class analysis cannot be overstated. By simultaneously incorporating individual-level and group-level variables, this technique reduces biases and improves classification accuracy compared to traditional single-level latent class models. This advancement in methodological rigor brings a substantially improved resolution in understanding adolescent substance use, enabling the disentangling of environmental context effects from individual behavioral tendencies.

In practical terms, the implications for mental health and addiction prevention are profound. The granular profile of substance use classes generated by this research provides a strategic roadmap for implementing differentiated prevention approaches. Instead of a one-size-fits-all tactic, resources can now be directed to address the unique constellation of risk factors and substance use patterns characteristic of each subgroup, potentially enhancing the effectiveness of early interventions and treatment outcomes.

Furthermore, the study’s embrace of multilevel modeling aligns with modern understandings of adolescent development as inherently embedded within ecological systems. Recognizing the dynamic interplay between individual dispositions and environmental contexts exemplifies a paradigm shift towards nuanced, evidence-driven public health strategies. It also underscores the importance of integrating wide-ranging data sources—ranging from census demographics to school environments—to build more holistic prevention frameworks.

Statistically, the application of MLCA in this context represents an innovative use of latent variable modeling, allowing researchers to infer unobserved heterogeneity from observable data streams. The complexity of adolescent substance use behavior necessitates such advanced analytical tools that can parse overlapping patterns which would otherwise be masked in aggregate data. This methodological innovation thus serves as a vital template for future research in related behavioral health domains.

The Canadian context adds particular salience given the country’s diverse population and varying provincial policies on substances like cannabis, which was recently legalized for recreational use. This heterogeneity in policy environments further accentuates the variability in adolescent use patterns, making rigorous multilevel approaches even more critical for accurate characterization and timely intervention.

Overall, this research marks a significant leap forward in our understanding of adolescent substance use, delivering a high-resolution map of the behavioral landscape that transcends simplistic characterizations. By employing cutting-edge statistical modeling and drawing upon a vast, representative sample, Coelho and colleagues have set a new benchmark for epidemiologic inquiries into youth substance behaviors.

This study does not merely illuminate patterns, but it invites a rethinking of intervention design, resource allocation, and even future data collection protocols. Its comprehensive framework provides a powerful tool to dissect complexity, facilitate targeted public health responses, and ultimately contribute to reducing the burden of substance use disorders among youth.

The richness of the findings encourages further exploration into the longitudinal stability and transition dynamics between latent classes, which could reveal how adolescent substance behaviors evolve over time and influence long-term mental health outcomes. Such knowledge would be invaluable for crafting dynamic prevention strategies responsive to the shifting realities of adolescence.

In summation, the multilevel latent class analysis approach championed in this work exemplifies a sophisticated, rigorous, and socially attuned scientific method. By embracing heterogeneity rather than searching for one-size-fits-all answers, this research epitomizes the future of substance use epidemiology—a future that promises more precision, inclusivity, and ultimately, impact.

With substance use among adolescents posing a persistent public health challenge globally, studies like this offer a beacon of hope. Through meticulous analytics and comprehensive population data, actionable insights emerge, fostering a more informed, targeted, and effective approach to safeguarding youth well-being in a complex world.

Subject of Research: Characterizing heterogeneity in adolescent substance use through multilevel latent class analysis.

Article Title: Characterizing Heterogeneity in Substance Use in a Large Sample of Canadian Adolescents: A Multilevel Latent Class Analysis.

Article References:
Coelho, S.G., Wardell, J.D., Flora, D.B. et al. Characterizing Heterogeneity in Substance Use in a Large Sample of Canadian Adolescents: A Multilevel Latent Class Analysis. International Journal of Mental Health and Addiction (2025). https://doi.org/10.1007/s11469-025-01586-z

Image Credits: AI Generated

DOI: https://doi.org/10.1007/s11469-025-01586-z