New approaches for examining associations with latent categorical variables: applications to substance abuse and aggression

A Latent Profile Analysis of Social Support, Online Contacts, and Preference for Online Communication Among Same- and Both-Sex Attracted and Other-Sex Attracted Adolescents

This study aimed to assess differences between other-sex attracted and same- and both-sex attracted adolescents in profiles of peer and family social support, online contacts, and preferences for online communication. Data stem from the 2017 Dutch "Health and Behavior in School-Aged Children" (HBSC) survey (N = 6,823; 4.0% same- and both-sex attracted; M age=14.73, SD = 1.59, range = 12-18). We conducted latent profile analyses to estimate profiles in peer and family social support, online contacts, and preferences for online communication. Then we assessed the association between sexual attraction and profile membership. A five-profile solution fitted the data best. Profiles were characterized as high support, online contact, and average online communication preference (35.6%); high support, low online contact, and weak online communication preference (42.9%); average support, high online contact, and strong online communication preference (9.9%); low support, low online contact, and average online communication preference (6.9%); and low support, average online contact, and average online communication preference (5.0%). Same- and both-sex attracted adolescents had higher odds than other-sex attracted adolescents of being in the latter three profiles than in the first profile. Thus, same- and both-sex attracted adolescents were more likely to report average to low rates of peer and family social support, high to low frequency of online contact, and an average to strong preference for online communication than other-sex attracted adolescents. The average to low levels of support especially influenced these sexual orientation-based differences in profile membership.

Treatment of worry and comorbid symptoms within depression, anxiety, and insomnia with a group-based rumination-focused cognitive-behaviour therapy in a primary health care setting: a randomised controlled trial

Introduction

Repetitive negative thinking (RNT) has been described as a maintaining transdiagnostic factor for psychopathology within the areas of depression, anxiety and insomnia. We investigated the effects of rumination-focused cognitive-behaviour therapy (RF-CBT) in a group format at a primary health care centre on symptoms of depression, anxiety, insomnia, RNT, and quality of life. The participants presented clinical symptom levels of worry and at least two disorders among anxiety disorders, major depressive disorder, and insomnia disorder.

Methods

A randomised controlled superiority parallel arm trial was used. 73 participants were included and randomised in pairs to either group-administered RF-CBT or a waiting list condition. The primary outcomes were self-rated worry and transdiagnostic symptoms (depression, anxiety, and insomnia). Intention-to-treat analyses of group differences were conducted using linear mixed models. Adverse side effects and incidents were presented descriptively.

Results

Group RF-CBT significantly reduced self-reported insomnia at post-treatment and self-reported insomnia and depression at the 2 month-follow-up, relative to the wait-list control group. There was no significant difference in change in RNT, anxiety, or quality of life.

Discussion

The current study suggests that group-administered RF-CBT may be effective for insomnia and potentially effective for depression symptomatology. However, the study was underpowered to detect small and moderate effects and the results should therefore be interpreted with caution.

Associations of Relationship Experiences, Dating Violence, Sexual Harassment, and Assault With Alcohol Use Among Sexual and Gender Minority Adolescents

Sexual and gender minority (SGM) adolescents report higher rates of dating violence victimization compared with their heterosexual and cisgender peers. Research on dating violence often neglects diversity in sexual and gender identities and is limited to experiences in relationships. Further, given that dating violence and alcohol use are comorbid, research on experiences of dating violence could provide insights into alcohol use disparities among SGM adolescents. We aimed to map patterns of relationship experiences, sexual and physical dating violence, and sexual and physical assault and explored differences in these experiences among SGM adolescents. Further, we examined how these patterns explained alcohol use. We used a U.S. non-probability national web-based survey administered to 13-17-year-old SGM adolescents (N = 12,534). Using latent class analyses, four patterns were identified: low relationship experience, dating violence and harassment and assault (72.0%), intermediate dating experiences, sexual harassment, and assault and low levels of dating violence (13.1%), high dating experiences, dating violence, and sexual assault (8.6%), and high dating experiences, dating violence, and sexual harassment and assault (6.3%). Compared to lesbian and gay adolescents, bisexual adolescents reported more experiences with dating, dating violence, and sexual assault, whereas heterosexual adolescents reported fewer experiences with dating, dating violence, and sexual harassment and assault. Compared to cisgender boys, cisgender girls, transgender boys, and non-binary/assigned male at birth adolescents were more likely to experience dating violence inside and outside of relationship contexts. Experiences of dating, dating violence, and sexual harassment and assault were associated with both drinking frequency and heavy episodic drinking. Together, the findings emphasize the relevance of relationship experiences when studying dating violence and how dating violence and sexual harassment and assault might explain disparities in alcohol use.

A Latent Class Analysis of Sexual Behavior and Associations with Sex Education, Smoking, Drinking, and Pornography Use Among Chinese Youth

The incidence of sexually transmitted diseases has alarmingly increased among Chinese youth in the past decade, and newly diagnosed cases of AIDS have almost doubled between 2010 and 2015. However, little is known about classes of sexual behavior among Chinese youth and associations with their other health behavior or experience of sex education. This study aimed to first identify classes of sexual behavior using latent class analysis and then to examine their associations with sex education, smoking, alcohol drinking, and pornography use. Data were from 13,865 unmarried youth aged 18-24 from the 2009 National Youth Reproductive Health Survey. Six indicators were used: timing of the first sexual intercourse, the number of past-year sexual partners, sex outside of a relationship, incidence of pregnancy, and contraceptive use at the latest and the first sexual intercourse. Four classes were identified: no sex (Class 1, 69%), safer sex (Class 2, 13%), early risk (Class 3, 13%), and multiple risk (Class 4, 5%). Smoking, drinking, and regular pornography use were associated with increased odds of being in the multiple risk class relative to the safer sex class. Sex education, smoking, and pornography use were associated with decreased odds of being in the no sex class relative to the safer sex class. The findings revealed qualitatively different classes of sexual behavior among Chinese youth and important roles of sex education, substance use, and pornography use. Reproductive health service providers can offer individually tailored services to serve youth with different profiles and needs.

Emotional and behavioural pathways to adolescent substance use and antisocial behaviour: results from the UK Millennium Cohort Study

This study examines the emotional and behavioural pathways to adolescent substance use and antisocial behaviour. Using a sample of 17,223 participants from the UK Millennium Cohort Study, we applied parallel-process growth mixture modelling on emotional and behavioural symptoms in those aged 3-14 and employed latent class analysis to identify patterns of substance use and antisocial behaviours at age 14. We then performed a multinomial regression analysis to explore the association between emotional and behavioural trajectories and patterns of adolescent substance use and antisocial behaviours, including sociodemographic, family, and maternal factors. We found five trajectories of emotional and behavioural symptoms and four classes of adolescence substance use and antisocial behaviour. Children and adolescents in the 'high externalising and internalising' and 'moderate externalising' trajectories were more likely to belong to any problematic behaviour class, especially the 'poly-substance use and antisocial behaviours' class. Inclusion in the 'moderate externalising and internalising (childhood limited)' class was associated with higher odds of belonging to the 'alcohol and tobacco' class. These associations remained significant after adjusting for important sociodemographic and contextual factors, such as maternal substance use, poverty, and parental status. Interventions on adolescent health promotion and risk behaviour prevention need to address the clustering of substance use and antisocial behaviour as well as the significant influence of early and chronic internalising and externalising symptoms on the aetiology of these behaviours.

Adolescent E-cigarette Users at Highest Risk of Cigarette Smoking Intention

Objectives: Adolescent e-cigarette users are at risk of developing smoking intention, an established predictor of conventional cigarette smoking. In this study, we identify subgroups of adolescent e-cigarette users who are most likely to intend to smoke conventional cigarettes. Methods: Cross-sectional data on 1357 8th and 10th grade e-cigarette users who had never smoked conventional cigarettes were obtained from 2014-2017 Monitoring the Future Surveys. We conducted latent class analysis to identify subgroups of adolescent e-cigarette users; through latent class regression analysis, we examined the association between subgroup membership and smoking intention. Results: We identified 3 subgroups of adolescent e-cigarette users: socially-protected (56.6%), peer-driven (29.8%), and market-vulnerable (13.6%). The peer-driven class reported the highest number of peers who smoke and the lowest proportion of friends who strongly disapproved of daily cigarette smoking. They were significantly more likely than the socially-protected and market-vulnerable classes to have smoking intention (AOR=2.46; 95% CI 1.84-3.28, and AOR=2.29; 95% CI 1.48-3.53, respectively). Conclusion: Our findings provide insights on the constellation of risk and protective factors that contribute to smoking intention among adolescent e-cigarette users. It highlights peer influence as an important area of emphasis for adolescent smoking prevention programs.

Profiles and Predictors of Dating Violence Among Sexual and Gender Minority Adolescents

PURPOSE

Sexual and gender minority adolescents report higher levels of dating violence compared with their heterosexual and cisgender peers. The objectives of the present study were to (1) identify latent profiles of dating violence; (2) examine if sexual and gender minority adolescents were particularly vulnerable to certain profiles of dating violence; and (3) explore how experiences of peer victimization, discrimination, and parental maltreatment explained this greater vulnerability.

METHODS

High school students in Grades 9 and 11 from the 2016 Minnesota Student Survey (N = 87,532; mean age = 15.29 years, SD = 1.23) were asked about their sexual and gender identities, their gender nonconformity, their experiences of verbal, physical, and sexual dating violence victimization and perpetration, as well their experiences of childhood maltreatment, peer victimization, and gender-based and sexual minority status-based discrimination.

RESULTS

Multinomial logistic regression analysis in a three-step latent class analysis procedure suggested five profiles of dating violence victimization and perpetration across the entire sample. Sexual and gender minority adolescents were generally more likely to be in classes high in dating violence victimization, perpetration, or both, compared with their heterosexual and cisgender peers. Gender nonconformity was also associated with greater risk for being in high dating violence classes. These differences, however, were generally nonsignificant when the social stressors of childhood maltreatment, peer victimization, and experiences of discrimination were accounted for.

CONCLUSIONS

Although findings suggested greater vulnerability for dating violence among sexual and gender minority adolescents, they underscore the importance of how minority stressors generally accounted for this greater vulnerability for dating violence.

Effect of Parameterization on Statistical Power and Effect Size Estimation in Latent Growth Modeling

The difference between groups in their random slopes is frequently examined in latent growth modeling to evaluate treatment efficacy. However, when end centering is used for model parameterization with a randomized design, the difference in the random intercepts is the model-estimated mean difference between the groups at the end of the study, which has the same expected value as the product of the coefficient for the slope difference and study duration. A Monte Carlo study found that (a) the statistical power to detect the treatment effect was greater when determined from the intercept instead of the slope difference, and (b) the standard error of the model-estimated mean difference was smaller when obtained from the intercept difference. Investigators may reduce Type II errors by comparing groups in random intercepts instead of random slopes to test treatment effects, and should therefore conduct power assessments using end centering to detect each difference.

Association between aggressive and non-fatal suicidal behaviors among U.S. high school students

OBJECTIVE

This study quantified the association between aggressive and non-fatal suicidal behaviors (NFSB) among U.S. high school students and examined whether the association could be explained by substance use, bullying and sexual/dating violence victimization, and other potential risk factors.

METHOD

Data were based on self-reports from 14,765 students who responded to the 2017 National Youth Risk Behavior Survey. Confirmatory latent class analysis (LCA) identified two distinct, dichotomous latent class variables manifested by indicators of past-year NFSB (i.e., ideation, plan, and attempt) and aggressive behavior (i.e., physical fighting in general and on school property). The structural model estimated the odds ratios between NFSB, aggressive behavior, and their potential risk factors.

RESULTS

Without adjusting for covariates, the confirmatory LCA estimated an odds ratio (OR) of 2.55 (95% confidence interval [CI]: 1.93, 3.37) between two latent class variables for violence against self (NFSB) and others (physical fighting). The net association, however, was rendered nonsignificant (OR = 1.08 [95% CI: 0.88, 1.31]) when adjusted for covariates. Significant common risk factors included exposure to physical dating violence, being bullied on school property and/or electronically, being threatened or injured by someone with a weapon on school property, and lifetime illegal drug use and prescription opioid misuse.

LIMITATIONS

Cross-sectional data do not allow assessment of causal relationships.

CONCLUSIONS

Findings elucidated the association between NFSB and aggressive behavior, with serious implications for prevention and intervention. Targeting substance use, bullying, and sexual and dating violence will protect students from engaging in both types of violent behaviors.

Latent classes of lifetime drug use disorder in national epidemiological survey on alcohol and related conditions - III

AIMS/BACKGROUND

People diagnosed with multiple drug use disorders are high-risk subpopulations, but changes in diagnostic classification and drug use prevalence mean patterning of drug use disorders has changed in the past decade. We analyzed comorbidity patterns of lifetime drug use disorder in a general population sample.

DESIGN

Using latent class analysis, we derived lifetime drug use disorder classes based on dichotomous indicators of sedative, cannabis, opioid, cocaine, stimulant, hallucinogen, inhalant/solvent, club drug, heroin, and other drug use disorders in the National Epidemiologic Survey on Alcohol and Related Conditions-III (n = 36,309). Multinomial models assessed associations between sociodemographic and clinical correlates and latent class status.

RESULTS

Four latent classes of lifetime substance use disorder were identified: A very low risk class, a class with high opioid, sedative and heroin comorbidity, a class based on cocaine and stimulant comorbidity, and class with high likelihood of multiple lifetime drug use disorders. All higher risk classes were associated with higher risk of lifetime personality disorder and mood disorder. Conduct disorder was also associated with higher risk, but level of risk varied by class. Opioid and sedative class was associated with higher odds of lifetime eating disorder diagnosis.

CONCLUSIONS

Comorbidity of drug use disorders is associated with a range of lifetime mental health disorder diagnoses. Unlike previous research, we did not identify a cannabis use class, possibly due to changes in diagnostic criteria and cannabis prevalence rates.

Social Stability Relates Social Conditions to the Syndemic of Sex, Drugs, and Violence

The distribution of violence, sexually transmitted infections, and substance use disorders is not random, but rather the product of disease, behavior, and social conditions that co-occur in synergistic ways (syndemics). Syndemics often disproportionately affect urban communities. Studies of syndemics, however, rarely apply consistent measures of social conditions. Here, the construct of social stability (SS) (housing, legal, residential, income, employment, and relationship stability) was evaluated as a consistent measure of social conditions related to sex, drug, and violence exposures in a new population in a Mid-Atlantic urban center. Lower SS predicted greater likelihood of any and combinations of risk. The magnitude varied based on specification: odds of sex-drug-violence exposure were greater for low vs. high latent SS class (OR = 6.25; 95%CI = 2.46, 15.96) compared with low vs. high SS category (OR = 2.64; 95%CI = 1.29, 5.39). A latent class characterized by residential instability was associated with greater likelihood of risk-a relationship that would have been missed with SS characterized only as an ordinal category. SS reliably captured social conditions associated with sexual, drug, and violence risks, and both quantity and quality of SS matter.

Latent Classes of Sexual Risk Behavior and Engagement in Outreach, Intervention and Prevention Services Among Women Who Inject Drugs Across 20 US Cities

BACKGROUND

Monitoring the effects of HIV prevention efforts among persons who inject drugs is key to informing prevention programs and policy.

METHODS

Data for this study came from the 2012 National HIV Behavioral Surveillance interviews with persons who inject drugs across 20 US cities. The present analyses include those who identified as female, ever had sex with a man, and were at risk of HIV infection (did not report a previous positive HIV test result) (n = 2624). We conducted latent class analysis to identify sexual risk classes, and modeled associations with engagement in HIV prevention services and HIV test results.

RESULTS

We identified 6 classes of sexual risk behavior: (1) low risk, (2) monogamous, (3) casual partner, (4) multiple partners, (5) exchange sex, and (6) exchange plus main partner. The class distribution was similar across the mainland regions. Bisexual orientation and homelessness were significant predictors of higher-risk class. HIV prevalence and participation in behavioral interventions did not vary significantly by risk class, while obtaining and using free condoms did. Independent of risk class, women in cities in the South were significantly less likely to use free condoms, and HIV prevalence was higher among non-Hispanic black women and women aged 40-49 years.

CONCLUSIONS

Bisexual orientation and homelessness were predictors of higher risk. Condom distribution programs reached fewer women in cities in the South. Race and age disparities in HIV-positive rates persisted after adjusting for sexual risk class.

Trust in genomic data sharing among members of the general public in the UK, USA, Canada and Australia

Trust may be important in shaping public attitudes to genetics and intentions to participate in genomics research and big data initiatives. As such, we examined trust in data sharing among the general public. A cross-sectional online survey collected responses from representative publics in the USA, Canada, UK and Australia (n = 8967). Participants were most likely to trust their medical doctor and less likely to trust other entities named. Company researchers were least likely to be trusted. Low, Variable and High Trust classes were defined using latent class analysis. Members of the High Trust class were more likely to be under 50 years, male, with children, hold religious beliefs, have personal experience of genetics and be from the USA. They were most likely to be willing to donate their genomic and health data for clinical and research uses. The Low Trust class were less reassured than other respondents by laws preventing exploitation of donated information. Variation in trust, its relation to areas of concern about the use of genomic data and potential of legislation are considered. These findings have relevance for efforts to expand genomic medicine and data sharing beyond those with personal experience of genetics or research participants.

Gender-specific substance use patterns and associations with individual, family, peer, and school factors in 15-year-old Portuguese adolescents: a latent class regression analysis

BACKGROUND

Adolescence is a critical period of vulnerability to substance use. Recent research has shown that gender differences in adolescence substance use are complex and in constant flux. The present study aims to investigate gender differences in substance use and initiation patterns in male and female adolescents, and to assess individual, family, peer, and school associated factors of these patterns.

METHODS

We applied latent class regression analysis to a Portuguese representative population sample of 1551 15-year-old adolescents, drawn from the 2010 'Health Behavior in School-Aged Children' survey, to characterise different profiles of substance use and initiation for boys and girls, and to identify factors associated with latent class membership, stratifying the associations analysis by gender.

RESULTS

Three common classes were found for both genders, specifically, Non-Users (boys [B] 34.42%, girls [G] 26.79%), Alcohol Experimenters (B 38.79%, G 43.98%) and Alcohol and Tobacco Frequent Users (B 21.31%, G 10.36%), with two additional unique classes: Alcohol Experimenters and Tobacco Users in girls (18.87%), and Early Initiation and Poly-Substance Users in boys (5.48%). Poor school satisfaction, bullying, fighting and higher family affluence scale score formed a common core of associated factors of substance use, although we found gender differences in these associations. In girls, but not in boys, family factors were associated with more problematic substance use. Not living with both parents was associated with girl's Alcohol and Tobacco Frequent Users (gATFU) class (OR 3.78 CI 1.18-12.11) and Alcohol Experimenters and Tobacco Users (AETU) class (OR 3.22 CI 1.4-7.44). Poor communication with mother was also associated with gATFU class membership (OR 3.82 CI 1.26-11.53) and AETU class (OR 3.66 CI 1.99-6.75). Additionally, a higher psychological symptoms score was associated with gATFU class membership (OR 1.16 CI 1.02-1.31).

CONCLUSION

Although we found common patterns and associated factors between boys and girls, we report two unique patterns of substance use in boys and girls and specific associations between family, school and peers, and individual factors with these patterns. These findings underscore the need for substance use prevention and health promotion programmes that address potential differences in substance use patterns and associated factors.

Outcomes related to adverse childhood experiences in college students: Comparing latent class analysis and cumulative risk

The aims of this study were to identify latent classes of adverse childhood experiences (ACEs) in a large sample of college students (N = 8997), investigate the relations between ACEs classes and life functioning, and compare results using latent class analysis to analyses using cumulative risk scores. Nine types of ACEs were assessed (three types of child abuse and six types of household dysfunction). Outcomes were self-report measures of mental health, physical health, alcohol consequences, and academic performance. Latent class analysis (LCA) results indicated that four classes fit the data best across random halves of the sample and were labeled High ACEs, Moderate Risk of Non-Violent Household Dysfunction, Emotional and Physical Child Abuse, and Low ACEs. Comparing across latent classes, the largest differences in outcomes were between the High ACEs and Low ACEs classes. There were no differences in outcomes between the Moderate Risk of Non-Violent Household Dysfunction and Emotional and Physical Child Abuse classes. The largest between-class differences were found for mental health and the smallest differences were found for academic performance. Comparing results using LCA latent classes and cumulative ACEs scores, the differences between the High and Low ACEs latent classes were similar to the differences between those with zero ACEs and those with 5 or more ACEs. Both approaches also accounted for roughly equivalent amounts of variance in all outcomes. Thus, latent class and cumulative risk analyses provided similar results with regard to predicting outcomes of interest among college students.

Maltreatment in childhood and intimate partner violence: A latent class growth analysis in a South African pregnancy cohort

Intimate partner violence (IPV) is a significant global problem, prevalent in low and middle-income countries (LMICs). IPV is particularly problematic during the perinatal and early postnatal period, where it is linked with negative maternal and child health outcomes. There has been little examination of profiles of IPV and early life adversity in LMIC contexts. We aimed to characterize longitudinal IPV and to investigate maternal maltreatment in childhood as a predictor of IPV exposure during pregnancy and postnatally in a low resource setting. This study was nested in the Drakenstein Child Health Study, a longitudinal birth cohort. Maternal IPV (emotional, physical and sexual) was measured at six timepoints from pregnancy to two years postpartum (n = 832); sociodemographic variables and maternal maltreatment in childhood were measured antenatally at 28-32 weeks' gestation. Associations between maternal maltreatment in childhood and IPV latent class membership (to identify patterns of maternal IPV exposure) were estimated using multinomial and logistic regression. We observed high levels of maternal maltreatment during childhood (34%) and IPV during pregnancy (33%). In latent class analysis separating by IPV sub-type, two latent classes of no/low and moderate sexual IPV and three classes of low, moderate, and high emotional and physical IPV (separately) were detected. In combined latent class analysis, including all IPV sub-types together, a low, moderate and high exposure class emerged as well as a high antenatal/decreasing postnatal class. Moderate and high classes for all IPV sub-types and combined analysis showed stable intensity profiles. Maternal childhood sexual abuse, physical abuse and neglect, and emotional abuse predicted membership in high IPV classes, across all domains of IPV (aORs between 1.99 and 5.86). Maternal maltreatment in childhood was associated with increased probability of experiencing high or moderate intensity IPV during and around pregnancy; emotional neglect was associated with decreasing IPV class for combined model. Intervening early to disrupt this cycle of abuse is critical to two generations.

A Comparison of Mixture Modeling Approaches in Latent Class Models With External Variables Under Small Samples

The purpose of the present study is to compare performances of mixture modeling approaches (i.e., one-step approach, three-step maximum-likelihood approach, three-step BCH approach, and LTB approach) based on diverse sample size conditions. To carry out this research, two simulation studies were conducted with two different models, a latent class model with three predictor variables and a latent class model with one distal outcome variable. For the simulation, data were generated under the conditions of different sample sizes (100, 200, 300, 500, 1,000), entropy (0.6, 0.7, 0.8, 0.9), and the variance of a distal outcome (homoscedasticity, heteroscedasticity). For evaluation criteria, parameter estimates bias, standard error bias, mean squared error, and coverage were used. Results demonstrate that the three-step approaches produced more stable and better estimations than the other approaches even with a small sample size of 100. This research differs from previous studies in the sense that various models were used to compare the approaches and smaller sample size conditions were used. Furthermore, the results supporting the superiority of the three-step approaches even in poorly manipulated conditions indicate the advantage of these approaches.

Disparities for LGBTQ and Gender Nonconforming Adolescents

OBJECTIVES

To identify patterns of childhood adversity in a sample of adolescents and assess disparities in these experiences for lesbian, gay, bisexual, transgender, and questioning adolescents and by level of gender nonconformity.

METHOD

By using the cross-sectional, statewide, anonymous 2016 Minnesota Student Survey, 81 885 students were included in the current study (50.59% male; mean age = 15.51). Participants were enrolled in grades 9 and 11 in a total of 348 schools.

RESULTS

Four patterns of childhood adversity were identified with sex-stratified latent class analyses (entropy = 0.833 males; 0.833 females), ranging from relatively low levels of abuse (85.3% males; 80.1% females) to polyvictimization (0.84% males; 1.98% females). A regression analysis showed that compared with heterosexual adolescents, gay, lesbian, bisexual, and questioning adolescents were more likely to be classified into profiles characterized by polyvictimization (odds ratio [OR] 1.81-7.53) and psychological and/or physical abuse (OR 1.29-3.12), than no or low adversity. Similarly, compared with nontransgender adolescents, transgender adolescents were more likely to be classified into profiles characterized by patterns of polyvictimization (OR 1.49-2.91) and psychological and/or physical abuse (OR 1.23-1.96). A higher level of gender nonconformity predicted a higher likelihood of being classified into each adversity profile compared with the no or low adversity profile (OR 1.14-1.45).

CONCLUSIONS

Sexual minority adolescents and adolescents with high levels of gender nonconformity are vulnerable to experience adversity. The disparities for lesbian, gay, bisexual, transgender, and questioning adolescents and adolescents with high gender nonconformity highlight the variation in patterns of childhood adversity that these youth are at risk of experiencing. The findings reveal the need for further research on the benefits and harm of screening for childhood adversity by physicians and pediatricians.

Injection-Related Risk Behavior and Engagement in Outreach, Intervention and Prevention Services Across 20 US Cities

BACKGROUND

Monitoring the effects of HIV prevention efforts on risk behaviors among persons who inject drugs is a key to inform prevention programs and policy.

METHODS

Using data from the 2012 National HIV Behavioral Surveillance interviews with persons who inject drugs across 20 US cities (n = 10,171), we conducted latent class analysis to identify injection risk classes and assess the relationship between engagement in prevention services and injection-related risk behavior. We conducted stratified analyses to examine the consistency of these associations across different geographical regions.

RESULTS

The latent class analysis identified 6 distinct classes of injection-related risk behavior. The class structure was consistent across regions of the United States, but the distribution of risk classes varied significantly across regions. With covariate adjustment, the South had the most high-risk behavior (21%) and the Midwest had the least (6%). Participation in syringe access services and other prevention services was the lowest in the South. Syringe access was associated with a significantly lower likelihood of membership in the highest risk class in all regions except the Midwest. Participation in individual or group intervention with a practical skills component was associated with less risky injection behavior in all regions except the Northeast. Interventions that featured only safer injection information and discussion had no relationship with risk class.

CONCLUSIONS

Our findings support evidence of the effectiveness of syringe service programs and safer injection skills training in reducing high-risk injection behavior and underscore the need to improve access to these prevention interventions in the South of the United States.

Parental alcohol use and risk of behavioral and emotional problems in offspring

OBJECTIVE

The majority of studies that have examined parental alcohol use and offspring outcomes have either focused on exposure in the antenatal period or from clinical populations. This study sought to examine proximal and distal associations between parental alcohol use and offspring conduct problems and depressive symptoms in a population birth cohort.

METHODS

We used prospective data from a large UK based population cohort (ALSPAC) to investigate the association between parental alcohol use, measured in units, (assessed at ages 4 and 12 years) with childhood conduct trajectories, (assessed on six occasions from 4 to 13.5 years, n = 6,927), and adolescent depressive symptoms (assessed on four occasions from ~13 to ~18 years, n = 5,539). Heavy drinking was defined as ≥21 units per week in mothers and partners who drank 4+ units daily.

RESULTS

We found little evidence to support a dose response association between parental alcohol use and offspring outcomes. For example, we found insufficient evidence to support an association between maternal alcohol use at age 4 years and childhood conduct problems (childhood limited: OR = 1.00, 95% CI = .99, 1.01; adolescent onset: OR = 0.99, 95% CI = .98, 1.00; and early-onset persistent: OR = 0.99, 95% CI = .98, 1.00) per 1-unit change in maternal alcohol use compared to those with low levels of conduct problems. We also found insufficient evidence to support an association between maternal alcohol use at age 4 years and adolescent depressive symptoms (intercept: b = .001, 95% CI = -.01, .01, and slope: b = .003, 95% CI = -.03, .03) per 1-unit change in maternal alcohol use. Results remained consistent across amount of alcohol consumed (i.e., number of alcohol units or heavy alcohol use), parent (maternal self-reports or maternal reports of partner's alcohol use), and timing of alcohol use (assessed at age 4 or age 12 years).

CONCLUSIONS

There is no support for an association between parental alcohol use during childhood and conduct and emotional problems during childhood or adolescence.

Latent classes of polydrug and polyroute use and associations with human immunodeficiency virus risk behaviours and overdose among people who inject drugs in Tijuana, Baja California, Mexico

INTRODUCTION AND AIMS

Patterns of polydrug use among people who inject drugs (PWID) may be differentially associated with overdose and unique human immunodeficiency virus (HIV) risk factors. Subgroups of PWID in Tijuana, Mexico, were identified based on substances used, route of administration, frequency of use and co-injection indicators.

DESIGN AND METHODS

Participants were PWID residing in Tijuana age ≥18 years sampled from 2011 to 2012 who reported injecting an illicit substance in the past month (n = 735). Latent class analysis identified discrete classes of polydrug use characterised by 11 indicators of past 6 months substance use. Multinomial logistic regression examined class membership association with HIV risk behaviours, overdose and other covariates using an automated three-step procedure in mplus to account for classification error.

RESULTS

Participants were classified into five subgroups. Two polydrug and polyroute classes were defined by use of multiple substances through several routes of administration and were primarily distinguished from each other by cocaine use (class 1: 5%) or no cocaine use (class 2: 29%). The other classes consisted primarily of injectors: cocaine, methamphetamine and heroin injection (class 3: 4%); methamphetamine and heroin injection (class 4: 10%); and heroin injection (class 5: 52%). Compared with the heroin-only injection class, memberships in the two polydrug and polyroute use classes were independently associated with both HIV injection and sexual risk behaviours.

DISCUSSION AND CONCLUSIONS

Substance use patterns among PWID in Tijuana are highly heterogeneous, and polydrug and polyroute users are a high-risk subgroup who may require more tailored prevention and treatment interventions. [Meacham MC, Roesch SC, Strathdee SA, Lindsay S, Gonzalez-Zuniga P, Gaines TL. Latent classes of polydrug and polyroute use and associations with human immunodeficiency virus risk behaviours and overdose among people who inject drugs in Tijuana, Baja California, Mexico. Drug Alcohol Rev 2018;37:128-136].

Trajectories of remission and mortality over 13 years after intake to substance use treatment

BACKGROUND

Little is known about the relationships between long-term patterns of substance use and mortality risk among substance use disorder (SUD) patients.

OBJECTIVE

To determine distinct patterns of remission and relapse of SUD over time and examine their relationship with mortality.

METHODS

The study site was Kaiser Permanente of Northern California. Data for 997 adults who entered substance use treatment between 1994 and 1996 and 4,241 non-SUD patients with similar demographic distributions (35% women in both groups) were analyzed. Latent class growth analysis identified distinct remission trajectory groups over 13 years among SUD patients, and survival analyses were conducted to examine the risk of death between remission trajectory groups, and SUD and non-SUD patients within each remission trajectory group.

RESULTS

Three distinct remission trajectory groups were identified among SUD patients: 1) early relapse-low remission probabilities; 2) declining remission-decreasing remission probabilities; and 3) stable remission-stable remission probabilities across all time points. Among the SUD patients, the early relapse group had a higher risk of death than those stably remitted; stable and declining remission groups did not differ. Comparisons within each remission trajectory group showed that SUD patients in the early relapse and stable remission groups had higher risks of death compared with non-SUD patients; there were no differences within the declining group.

CONCLUSIONS

SUD patients in the stable remission group had lower survival rates compared with non-SUD patients. These findings underline the importance of continuously addressing healthcare needs of individuals with SUD, even in the presence of long-term remission.

Longitudinal Patterns of Stages of Change for Exercise and Lifestyle Intervention Outcomes: An Application of Latent Class Analysis with Distal Outcomes

Stages of change measure an individual's readiness to alter a health behavior. This study examined the latent longitudinal patterns of stages of change (SoC) for regular exercise over time among individuals participating in a lifestyle intervention project. It also investigated the association between the longitudinal patterns of SoC and intervention outcomes using a new statistical method to assess the relationship between latent class membership and distal outcomes. We analyzed data from the Special Diabetes Program for Indians Diabetes Prevention Program, a lifestyle intervention program to prevent diabetes among American Indians and Alaska Natives. Latent class analysis (LCA) was conducted to identify the longitudinal patterns of SoC for regular exercise reported at three time points. LCA with distal outcomes was performed to investigate the associations between latent class membership and behavioral changes after the intervention. The parameters and standard errors of the LCA with distal outcomes models were estimated using an improved three-step approach. Three latent classes were identified: Pre-action, Transition, and Maintenance classes. The Transition class, where stage progression occurred, had the greatest improvements in physical activity and weight outcomes at both time points post-baseline among female participants. It also had the largest improvements in weight outcomes among male participants. Furthermore, the Pre-action class had more attenuation in the improvements they had achieved initially than the other two classes. These findings suggest the potential importance of motivating participants to modify their readiness for behavioral change in future lifestyle interventions.

Marijuana use and service utilization among adolescents 7 years post substance use treatment

BACKGROUND

In an environment of increasingly liberal attitudes towards marijuana use and legalization, little is known about long-term trajectories of marijuana use among clinical samples of adolescents, and how these trajectories relate to health services utilization over time.

METHODS

Latent growth curve analysis was used to identify distinct trajectories of marijuana use in a clinical sample of adolescents (N=391) over 7 years post substance use treatment in an integrated health system. We examined psychiatric problems and polysubstance use associated with the identified trajectory groups using general linear models. Nonlinear mixed-effects logistic regressions were used to examine associations between health services use and the trajectory groups.

RESULTS

We identified three marijuana use trajectory groups: (1) Abstinent (n=117); (2) Low/Stable use (n=174); and (3) Increasing use (n=100). Average externalizing and anxiety/depression scores were significantly lower over time for the Abstinent group compared to the Increasing and Low/Stable groups. The Low Stable and the Increasing group had fewer psychiatric visits over time (p<0.05) and the Low/Stable group used more substance use treatment services over time compared with the Abstinent group (p<0.001).

CONCLUSIONS

Treated adolescents showed distinct marijuana use patterns, one of which indicated a group of adolescents at risk of increased use over time. These individuals have greater psychiatric and polysubstance use over time, but may not be accessing needed services.

Early life socioeconomic position and immune response to persistent infections among elderly Latinos

Persistent infections, such as cytomegalovirus (CMV), herpes simplex virus-1 (HSV-1), Helicobacter pylori (H. pylori), and Toxoplasma gondii (T. gondii), are common in the U.S. but their prevalence varies by socioeconomic status. It is unclear if early or later life socioeconomic position (SEP) is a more salient driver of disparities in immune control of these infections. Using data from the Sacramento Area Latino Study on Aging, we examined whether early or later life SEP was the strongest predictor of immune control later in life by contrasting two life course models, the critical period model and the chain of risk model. Early life SEP was measured as a latent variable, derived from parental education and occupation, and food availability. Indicators for SEP in later life included education level and occupation. Individuals were categorized by immune response to each pathogen (seronegative, low, medium and high) with increasing immune response representing poorer immune control. Cumulative immune response was estimated using a latent profile analysis with higher total immune response representing poorer immune control. Structural equation models were used to examine direct, indirect and total effects of early life SEP on each infection and cumulative immune response, controlling for age and gender. The direct effect of early life SEP on immune response was not statistically significant for the infections or cumulative immune response. Higher early life SEP was associated with lower immune response for T. gondii, H. pylori and cumulative immune response through pathways mediated by later life SEP. For CMV, higher early life SEP was both directly associated and partially mediated by later life SEP. No association was found between SEP and HSV-1. Findings from this study support a chain of risk model, whereby early life SEP acts through later life SEP to affect immune response to persistent infections in older age.

Alexithymia and psychosocial problems among Italian preadolescents. A latent class analysis approach

The study, conducted on Italian preadolscents aged 11 to 13 belonging to the general population, aims to investigate the relationship between the emotional functioning, namely, alexithymia, and the risk of developing behavioral and emotional problems measured using the Strength and Difficulty Questionnaire. The latent class analysis approach allowed to identify two latent variables, accounting for the internalizing (emotional symptoms and difficulties in emotional awareness) and for the externalizing problems (conduct problems and hyperactivity, problematic relationships with peers, poor prosocial behaviors and externally oriented thinking). The two latent variables featured two latent classes: the difficulty in dealing with problems and the strength to face problems that was representative of most of the healthy participants with specific gender differences. Along with the analysis of psychopathological behaviors, the study of resilience and strengths can prove to be a key step in order to develop valuable preventive approaches to tackle psychiatric disorders.

Patterns of Reasons for Taiwanese Adults' Health Information-Seeking Efforts: A Latent Class Analysis Approach

The aims of this study were to use latent class analysis (LCA) to identify subgroups of adults in Taiwan based on their reasons for seeking health information and to explore predictors of subgroup membership. A questionnaire survey of 752 adults from 25 communities in Taiwan was conducted. LCA was used to identify distinct classes of participants; latent class regression was performed to identify factors predicting latent class membership. Three classes emerged through LCA. The Health-Improving Group (50.40%) reported high probabilities of reasons relevant to improving their or someone else's health but low probabilities of reasons relevant to patient-provider interaction. The Active Group (32.98%) showed high probabilities of almost all of the reasons for seeking health information. The Passive Group (16.62%) showed low probabilities across all of the reasons. Compared to the Health-Improving Group, the Active Group was significantly more likely to have higher education and perceive higher information-seeking self-efficacy. The individuals in the Passive Group were significantly more likely to be male, be younger, have lower health literacy, and have fewer years of education than those in the Health-Improving Group. This LCA approach can provide important information on how communication strategies should be applied to different population subgroups.

Profiles of Recovery from Mood and Anxiety Disorders: A Person-Centered Exploration of People's Engagement in Self-Management

Context: A shift toward person-centered care has been occurring in services provided to people with mood and anxiety disorders. Recovery is recognized as encompassing personal aspects in addition to clinical ones. Guidelines now recommend supporting people's engagement in self-management as a complementary recovery avenue. Yet the literature lacks evidence on how individualized combinations of self-management strategies used by people relate to their clinical and personal recovery indicators.

Objectives: The aims of this study were to identify profiles underlying mental health recovery, describe the characteristics of participants corresponding to each profile, and examine the associations of profiles with criterion variables.

Method: 149 people recovering from anxiety, depressive, or bipolar disorders completed questionnaires on self-management, clinical recovery (symptom severity), personal recovery (positive mental health), and criterion variables (personal goal appraisal, social participation, self-care abilities, coping).

Results: Latent profile analysis (LPA) revealed three profiles. The Floundering profile included participants who rarely used self-management strategies and had moderately severe symptoms and the lowest positive mental health. The Flourishing profile was characterized by frequent use of self-empowerment strategies, the least severe symptoms, and the highest positive mental health. Participants in the Struggling profile engaged actively in several self-management strategies focused on symptom reduction and healthy lifestyle. They concomitantly reported high symptom severity and moderately high positive mental health. The study revealed that Floundering was associated with higher probabilities of being a man, being single, and having a low income. People in the Flourishing profile had the most favorable scores on criterion variables, supporting the profiles' construct validity.

Discussion: The mixed portrait of Struggling participants on recovery indicators suggests the relationship between health engagement and recovery is more intricate than anticipated. Practitioners should strive for a holistic understanding of their clients' self-management strategies and recovery indicators to provide support personalized to their profile. While people presenting risk factors would benefit from person-centered support, societal efforts are needed in the long term to reduce global health inequalities. The integration of constructs from diverse fields (patient-centered care, chronic illness, positive psychology) and the use of person-oriented analysis yielded new insights into people's engagement in their health and well-being.

Sociodemographic predictors of latent class membership of problematic and disordered gamblers

This paper reports a series of analyses examining the predictors of gambling subtypes identified from a latent class analysis of problem gambling assessment data, pooled from four health and gambling surveys conducted in Britain between 2007 and 2012. Previous analyses have indicated that gambling assessments have a consistent three class structure showing quantitative and potentially qualitative differences. Bringing this data together is useful for studying more severe problem gamblers, where the small number of respondents has been a chronic limitation of gambling prevalence research. Predictors were drawn from sociodemographic indicators and engagement with other legal addictive behaviours, namely smoking and alcohol consumption. The pooled data was entered into a multinomial logistic regression model in which class membership was regressed along a series of demographic variables and survey year, based on previous analyses of gambling prevalence data. The results identified multiple demographic differences (age, general health, SES, being single, membership of ethnic minority groups) between the non-problem and two classes endorsing some problem gambling indicators. Although these two groups tended to share a sociodemographic profile, the odds of being male, British Asian and a smoker increased between the three groups in line with problem gambling severity. Being widowed was also found to be associated with the most severe gambling class. A number of associations were also observed with other addictive behaviours. However these should be taken as indicative as these were limited subsamples of a single dataset. These findings identify specific groups in which gambling problems are more prevalent, and highlight the importance of the interaction between acute and determinant aspects of impulsivity, suggesting that a more complex account of impulsivity should be considered than is currently present in the gambling literature.

Heterogeneity in the Relationship of Substance Use to Risky Sexual Behavior Among Justice-Involved Youth: A Regression Mixture Modeling Approach

Justice-involved adolescents engage in high levels of risky sexual behavior and substance use, and understanding potential relationships among these constructs is important for effective HIV/STI prevention. A regression mixture modeling approach was used to determine whether subgroups could be identified based on the regression of two indicators of sexual risk (condom use and frequency of intercourse) on three measures of substance use (alcohol, marijuana and hard drugs). Three classes were observed among n = 596 adolescents on probation: none of the substances predicted outcomes for approximately 18 % of the sample; alcohol and marijuana use were predictive for approximately 59 % of the sample, and marijuana use and hard drug use were predictive in approximately 23 % of the sample. Demographic, individual difference, and additional sexual and substance use risk variables were examined in relation to class membership. Findings are discussed in terms of understanding profiles of risk behavior among at-risk youth.

Multifinality of peer victimization: maladjustment patterns and transitions from early to mid-adolescence

Peer victimization is a common and pervasive experience in childhood and adolescence and is associated with various maladjustment symptoms, including internalizing, externalizing, and somatic problems. This variety suggests that peer victimization is multifinal where exposure to the same risk leads to different outcomes. However, very little is known about the relative likelihood of each form of maladjustment. We used a latent profile approach to capture multiple possible outcomes and examined prediction by peer victimization. We also examined the role of peer victimization with regard to stability and change in maladjustment. Maladjustment symptoms and peer victimization were assessed from the participants of the large cohort study TRacking Adolescents' Individual Lives Survey in early and mid-adolescence. Latent profile and latent transition analyses were conducted to examine associations between victimization and maladjustment profile and to test the role of victimization in maladjustment profile transitions. Four maladjustment profiles were identified for early adolescence (Low, Internalizing, Externalizing, Comorbid) and three profiles (Low, Internalizing, Externalizing) were identified for mid-adolescence. Internalizing problems were more likely in victimized adolescents than low symptom levels or externalizing problems. Victimized adolescents were at greater risk to develop internalizing problems between early and mid-adolescence than non-victimized adolescents. Peer victimization is multifinal mostly when outcomes are examined separately. If multiple outcomes are tested simultaneously, internalizing problems seem to be the most likely outcome.

Latent Class Analysis of DSM-5 Alcohol Use Disorder Criteria Among Heavy-Drinking College Students

The DSM-5 has created significant changes in the definition of alcohol use disorders (AUDs). Limited work has considered the impact of these changes in specific populations, such as heavy-drinking college students. Latent class analysis (LCA) is a person-centered approach that divides a population into mutually exclusive and exhaustive latent classes, based on observable indicator variables. The present research was designed to examine whether there were distinct classes of heavy-drinking college students who met DSM-5 criteria for an AUD and whether gender, perceived social norms, use of protective behavioral strategies (PBS), drinking refusal self-efficacy (DRSE), self-perceptions of drinking identity, psychological distress, and membership in a fraternity/sorority would be associated with class membership. Three-hundred and ninety-four college students who met DSM-5 criteria for an AUD were recruited from three different universities. Two distinct classes emerged: Less Severe (86%), the majority of whom endorsed both drinking more than intended and tolerance, as well as met criteria for a mild AUD; and More Severe (14%), the majority of whom endorsed at least half of the DSM-5 AUD criteria and met criteria for a severe AUD. Relative to the Less Severe class, membership in the More Severe class was negatively associated with DRSE and positively associated with self-identification as a drinker. There is a distinct class of heavy-drinking college students with a more severe AUD and for whom intervention content needs to be more focused and tailored. Clinical implications are discussed.

Religiousness and Levels of Hazardous Alcohol Use: A Latent Profile Analysis

Prior person-centered research has consistently identified a subgroup of highly religious participants that uses significantly less alcohol when compared to the other subgroups. The construct of religious motivation is absent from existing examinations of the nuanced combinations of religiousness dimensions within persons, and alcohol expectancy valuations have yet to be included as outcome variables. Variable-centered approaches have found religious motivation and alcohol expectancy valuations to play a protective role against individuals' hazardous alcohol use. The current study examined latent religiousness profiles and hazardous alcohol use in a large, multisite sample of ethnically diverse college students. The sample consisted of 7412 college students aged 18-25 (M age = 19.77, SD age = 1.61; 75% female; 61% European American). Three latent profiles were derived from measures of religious involvement, salience, and religious motivations: Quest-Intrinsic Religiousness (highest levels of salience, involvement, and quest and intrinsic motivations; lowest level of extrinsic motivation), Moderate Religiousness (intermediate levels of salience, involvement, and motivations) and Extrinsic Religiousness (lowest levels of salience, involvement, and quest and intrinsic motivations; highest level of extrinsic motivation). The Quest-Intrinsic Religiousness profile scored significantly lower on hazardous alcohol use, positive expectancy outcomes, positive expectancy valuations, and negative expectancy valuations, and significantly higher on negative expectancy outcomes, compared to the other two profiles. The Extrinsic and Moderate Religiousness profiles did not differ significantly on positive expectancy outcomes, negative expectancy outcomes, negative expectancy valuations, or hazardous alcohol use. The results advance existing research by demonstrating that the protective influence of religiousness on college students' hazardous alcohol use may involve high levels on both quest and intrinsic religious motivation.

Confidence interval estimation for standardized effect sizes in multilevel and latent growth modeling

OBJECTIVE

Multilevel and latent growth models are frequently used interchangeably to examine differences between groups in trajectories of outcomes from controlled clinical trials. The unstandardized coefficient for the effect from group to slope (the treatment effect) from such models can be converted to a standardized mean difference (Cohen's d) between the treatment and control groups at end of study. This article addresses the confidence interval (CI) for this effect size.

METHOD

Two sets of equations for estimating the CI for the treatment effect size in multilevel models were derived, and the usage of each was illustrated with data from the National Youth Study (Elliott, Huizinga, & Menard, 1989). Validity of the CIs was examined with a Monte Carlo simulation study that manipulated effect potency and sample size.

RESULTS

The equivalence of the 2 new CI estimation methods was demonstrated, and the Monte Carlo study found that bias in the CI for the effect size was not appreciably larger than bias in the CI for the widely used unstandardized coefficient.

CONCLUSIONS

Investigators reporting this increasingly popular effect size can estimate its CI with equations presented in this article.

ADDRESSING CONFOUNDING WHEN ESTIMATING THE EFFECTS OF LATENT CLASSES ON A DISTAL OUTCOME

Confounding is widely recognized in settings where all variables are fully observed, yet recognition of and statistical methods to address confounding in the context of latent class regression are slowly emerging. In this study we focus on confounding when regressing a distal outcome on latent class; extending standard confounding methods is not straightforward when the treatment of interest is a latent variable. We describe a recent 1-step method, as well as two 3-step methods (modal and pseudoclass assignment) that incorporate propensity score weighting. Using simulated data, we compare the performance of these three adjusted methods to an unadjusted 1-step and unadjusted 3-step method. We also present an applied example regarding adolescent substance use treatment that examines the effect of treatment service class on subsequent substance use problems. Our simulations indicated that the adjusted 1-step method and both adjusted 3-step methods significantly reduced bias arising from confounding relative to the unadjusted 1-step and 3-step approaches. However, the adjusted 1-step method performed better than the adjusted 3-step methods with regard to bias and 95% CI coverage, particularly when class separation was poor. Our applied example also highlighted the importance of addressing confounding - both unadjusted methods indicated significant differences across treatment classes with respect to the outcome, yet these class differences were not significant when using any of the three adjusted methods. Potential confounding should be carefully considered when conducting latent class regression with a distal outcome; failure to do so may results in significantly biased effect estimates or incorrect inferences.

Associations of Women's Substance Dependency Symptoms with Intimate Partner Violence

Associations of substance dependencies and experiences with intimate partner violence (IPV) were examined in a community sample of 146 female participants in a longitudinal study of couples. The women with a history of dependence on hard drugs (but not alcohol, cannabis, or sedatives) were more likely to also have perpetrated IPV. However, only women dependent on cocaine were more likely to have been a victimized by their male partners. Psychological IPV was found to be more stable across time than physical IPV, but associations of substance abuse with IPV did not vary by IPV type. Findings were compared with results from a prior study of men's substance abuse and IPV that also found associations between dependence on hard drugs (but not alcohol dependence alone) and perpetration of IPV.

Pitfalls in subgroup analysis based on growth mixture models: a commentary on Van Leeuwen et al. (2012)

OBJECTIVES

This article is a brief commentary in response to "van Leeuwen et al. (Qual Life Res 21:1499-1508, 2012)"

METHODS AND RESULTS

The commentary argues that in the context of mixture modeling, assigning individuals to specific subgroups for conducting a secondary set of analyses ignores the original uncertainty in group membership, thereby biasing any subsequent results and inference.

CONCLUSIONS

Alternative approaches to subgroup analysis that attempt to preserve uncertainty in group membership are discussed and illustrated.