Inverse Propensity Score Weighting with a Latent Class Exposure: Estimating the Causal Effect of Reported Reasons for Alcohol Use on Problem Alcohol Use 16 Years Later

Unreliable Continuous Treatment Indicators in Propensity Score Analysis

Propensity score analyses (PSA) of continuous treatments often operationalize the treatment as a multi-indicator composite, and its composite reliability is unreported. Latent variables or factor scores accounting for this unreliability are seldom used as alternatives to composites. This study examines the effects of the unreliability of indicators of a latent treatment in PSA using the generalized propensity score (GPS). A Monte Carlo simulation study was conducted varying composite reliability, continuous treatment representation, variability of factor loadings, sample size, and number of treatment indicators to assess whether Average Treatment Effect (ATE) estimates differed in their relative bias, Root Mean Squared Error, and coverage rates. Results indicate that low composite reliability leads to underestimation of the ATE of latent continuous treatments, while the number of treatment indicators and variability of factor loadings show little effect on ATE estimates, after controlling for overall composite reliability. The results also show that, in correctly specified GPS models, the effects of low composite reliability can be somewhat ameliorated by using factor scores that were estimated including covariates. An illustrative example is provided using survey data to estimate the effect of teacher adoption of a workbook related to a virtual learning environment in the classroom.

Estimating the Effects of a Complex, Multidimensional Moderator: An Example of Latent Class Moderation to Examine Differential Intervention Effects of Substance Use Services

Improvements in substance use disorder recovery may be achieved by recognizing that effective interventions do not work equally well for all individuals. Heterogeneity of intervention effects is traditionally examined as a function of a single variable, such as gender or baseline severity. However, responsiveness to an intervention is likely a result of multiple, intersecting factors. Latent class moderation enables the examination of heterogeneity in intervention effects across subgroups characterized by profiles of characteristics. This study analyzed data from adolescents (aged 13 to 18 years old) who needed treatment for cannabis use (n = 14,854) and participated in the Global Appraisal of Individual Needs to evaluate differential effects of substance use services on cannabis use outcomes. We demonstrate an adjusted three-step approach using weights that account for measurement error; sample codes in Mplus and Latent Gold are provided and data are publicly available. Indicators of the latent class moderator comprised six contextual (e.g., recovery environment risk) and individual (e.g., internal mental distress) risk factors. The latent class moderator comprised four subgroups: low risk (21.1%), social risk (21.1%), environmental risk (12.5%), and mixed risk (45.2%). Limited moderation of associations between level of care and any past 90-day cannabis use were observed. In predicting number of cannabis use-days, compared to individuals with low risk, those with environmental risk showed improved outcomes from intensive outpatient care whereas individuals with social risk and mixed risk showed improved outcomes from residential care (all compared to early intervention/outpatient care). Latent class moderation holds potential to elucidate heterogeneity in intervention effectiveness that otherwise may go undetected.

Symptom patterns and health service use of women in early adulthood: a latent class analysis from the Australian Longitudinal Study on Women's Health

BACKGROUND

Symptoms can be strong drivers for initiating interaction with the health system, especially when they are frequent, severe or impact on daily activities. Research on symptoms often use counts of symptoms as a proxy for symptom burden, however simple counts don't provide information on whether groups of symptoms are likely to occur together or whether such groups are associated with different types and levels of healthcare use. Women have a higher symptom burden than men; however studies of symptom patterns in young women are lacking. We aimed to characterise subgroups of women in early adulthood who experienced different symptom patterns and to compare women's use of different types of health care across the different symptom subgroups.

METHODS

Survey and linked administrative data from 7 797 women aged 22-27 years in 2017 from the 1989-95 cohort of the Australian Longitudinal Study on Women's Health were analysed. A latent class analysis was conducted to identify subgroups of women based on the frequency of 16 symptom variables. To estimate the associations between the latent classes and health service use, we used the "Bolck, Croon and Hagenaars" (BCH) approach that takes account of classification error in the assignment of women to latent classes.

RESULTS

Four latent classes were identified, characterised by 1) low prevalence of most symptoms (36.6%), 2) high prevalence of menstrual symptoms but low prevalence of mood symptoms (21.9%), 3) high prevalence of mood symptoms but low prevalence of menstrual symptoms, (26.2%), and high prevalence of many symptoms (15.3%). Compared to the other three classes, women in the high prevalence of many symptoms class were more likely to visit general practitioners and specialists, use more medications, and more likely to have had a hospital admission.

CONCLUSIONS

Women in young adulthood experience substantially different symptom burdens. A sizeable proportion of women experience many co-occurring symptoms across both physical and psychological domains and this high symptom burden is associated with a high level of health service use. Further follow-up of the women in our study as they enter their late 20 s and early 30 s will allow us to examine the stability of the classes of symptoms and their associations with general health and health service use. Similar studies in other populations are needed to assess the generalisability of the findings.

Alcohol Use Disorder Symptoms Reported during Midlife: Results from the Monitoring the Future Study among US Adults at Modal Ages 50, 55, and 60

BACKGROUND

The extent to which adolescent substance use is associated with alcohol use disorder (AUD) symptoms in midlife is not yet fully explored.

METHODS

Longitudinal data from the national Monitoring the Future study was used. The sample included 11,830 12th graders (1976-1987) who were surveyed again at modal ages 50 (37.8%), 55 (46.3%), or 60 (15.8%) in 2008-2019. Approximately 48.7% were male; 81.5% identified as non-Hispanic White. Weighted logistic and multinomial logistic regressions were used to examine associations between past 30-day use of cigarettes, marijuana, and alcohol at age 18, sociodemographics, and a midlife AUD symptom outcome (coded as non-drinking, drinking without AUD [endorsed ≤1 criterion], or AUD symptoms [endorsed 2+ criteria]).

RESULTS

Prevalence of midlife AUD symptoms was 27.1%. Higher relative risk of reporting AUD symptoms (vs. drinking without AUD) was associated with age 18 substance use (any cigarette use [vs. no use], any marijuana use [vs. no use], binge drinking [vs. both no use and drinking at less than binge levels]), being male (vs. female), being non-Hispanic White (vs. non-Hispanic Black), and having a 4-year college degree. Higher relative risk of reporting non-drinking (vs. drinking without AUD) was associated with no 30-day alcohol use at age 18, being non-Hispanic Black or non-Hispanic other (vs. non-Hispanic White), and not having a 4-year college degree.

CONCLUSIONS

Findings suggest substance use at age 18 has meaningful associations with midlife AUD symptoms. Dissemination of prevention and intervention efforts in adolescence and early adulthood may be important for reducing hazardous midlife drinking.

Evaluating sensitivity to classification uncertainty in latent subgroup effect analyses

Background

Increasing attention is being given to assessing treatment effect heterogeneity among individuals belonging to qualitatively different latent subgroups. Inference routinely proceeds by first partitioning the individuals into subgroups, then estimating the subgroup-specific average treatment effects. However, because the subgroups are only latently associated with the observed variables, the actual individual subgroup memberships are rarely known with certainty in practice and thus have to be imputed. Ignoring the uncertainty in the imputed memberships precludes misclassification errors, potentially leading to biased results and incorrect conclusions.

Methods

We propose a strategy for assessing the sensitivity of inference to classification uncertainty when using such classify-analyze approaches for subgroup effect analyses. We exploit each individual’s typically nonzero predictive or posterior subgroup membership probabilities to gauge the stability of the resultant subgroup-specific average causal effects estimates over different, carefully selected subsets of the individuals. Because the membership probabilities are subject to sampling variability, we propose Monte Carlo confidence intervals that explicitly acknowledge the imprecision in the estimated subgroup memberships via perturbations using a parametric bootstrap. The proposal is widely applicable and avoids stringent causal or structural assumptions that existing bias-adjustment or bias-correction methods rely on.

Results

Using two different publicly available real-world datasets, we illustrate how the proposed strategy supplements existing latent subgroup effect analyses to shed light on the potential impact of classification uncertainty on inference. First, individuals are partitioned into latent subgroups based on their medical and health history. Then within each fixed latent subgroup, the average treatment effect is assessed using an augmented inverse propensity score weighted estimator. Finally, utilizing the proposed sensitivity analysis reveals different subgroup-specific effects that are mostly insensitive to potential misclassification.

Conclusions

Our proposed sensitivity analysis is straightforward to implement, provides both graphical and numerical summaries, and readily permits assessing the sensitivity of any machine learning-based causal effect estimator to classification uncertainty. We recommend making such sensitivity analyses more routine in latent subgroup effect analyses.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12874-022-01720-8.

Solitary alcohol use in adolescence predicts alcohol problems in adulthood: A 17-year longitudinal study in a large national sample of US high school students

BACKGROUND

Identifying risk factors for alcohol use disorder (AUD) is important for public health. The social context of drinking-such as drinking alone-may be an independent and robust early risk marker for AUD symptoms later in life. We evaluated whether solitary alcohol use in adolescence (age 18) and young adulthood (age 23/24) was concurrently associated with binge drinking and prospectively predicted age 35 AUD symptoms, and whether associations differed by sex.

METHODS

Longitudinal data were from the Monitoring the Future study. Surveys were completed by adolescents in 12th grade at age 18 (1976-2002), young adults at age 23/24 (1981-2008), and adults at age 35 (1993-2019). Analyses included past 12-month alcohol users (n = 4464 for adolescent models; n = 4561 for young adult models). Multivariable regression analyses tested whether adolescent and young adult solitary alcohol use was associated concurrently with binge drinking frequency and prospectively with age 35 AUD symptoms.

RESULTS

Solitary alcohol use in adolescence and young adulthood was associated (a) concurrently with binge drinking and (b) prospectively with increased risk of age 35 AUD symptoms (even after controlling for earlier binge drinking, alcohol use frequency, and sociodemographic covariates). Adolescent solitary alcohol use was associated with age 35 AUD symptoms particularly among females; no interaction was observed between sex and young adult solitary alcohol use in predicting age 35 AUD symptoms.

CONCLUSIONS

Adolescent and young adult solitary alcohol use was associated with increased adult AUD symptoms above and beyond other risk factors; adolescent female solitary alcohol users were especially at risk.

Sensitivity and specificity of information criteria

Information criteria (ICs) based on penalized likelihood, such as Akaike's information criterion (AIC), the Bayesian information criterion (BIC) and sample-size-adjusted versions of them, are widely used for model selection in health and biological research. However, different criteria sometimes support different models, leading to discussions about which is the most trustworthy. Some researchers and fields of study habitually use one or the other, often without a clearly stated justification. They may not realize that the criteria may disagree. Others try to compare models using multiple criteria but encounter ambiguity when different criteria lead to substantively different answers, leading to questions about which criterion is best. In this paper we present an alternative perspective on these criteria that can help in interpreting their practical implications. Specifically, in some cases the comparison of two models using ICs can be viewed as equivalent to a likelihood ratio test, with the different criteria representing different alpha levels and BIC being a more conservative test than AIC. This perspective may lead to insights about how to interpret the ICs in more complex situations. For example, AIC or BIC could be preferable, depending on the relative importance one assigns to sensitivity versus specificity. Understanding the differences and similarities among the ICs can make it easier to compare their results and to use them to make informed decisions.

Separating Algorithms From Questions and Causal Inference With Unmeasured Exposures: An Application to Birth Cohort Studies of Early Body Mass Index Rebound

Observational studies reporting on adjusted associations between childhood body mass index (BMI; weight (kg)/height (m)2) rebound and subsequent cardiometabolic outcomes have often not paid explicit attention to causal inference, including definition of a target causal effect and assumptions for unbiased estimation of that effect. Using data from 649 children in a Boston, Massachusetts-area cohort recruited in 1999-2002, we considered effects of stochastic interventions on a chosen subset of modifiable yet unmeasured exposures expected to be associated with early (<age 4 years) BMI rebound (a proxy measure) on adolescent cardiometabolic outcomes. We considered assumptions under which these effects might be identified with available data. This leads to an analysis where the proxy, rather than the exposure, acts as the exposure in the algorithm. We applied targeted maximum likelihood estimation, a doubly robust approach that naturally incorporates machine learning for nuisance parameters (e.g., propensity score). We found a protective effect of an intervention that assigns modifiable exposures according to the distribution in the observational study of persons without (vs. with) early BMI rebound for fat mass index (fat mass (kg)/ height (m)2; -1.39 units, 95% confidence interval: -1.63, -0.72) but weaker or no effects for other cardiometabolic outcomes. Our results clarify distinctions between algorithms and causal questions, encouraging explicit thinking in causal inference with complex exposures.

Drinking Intensity at Age 29/30 as a Predictor of Alcohol Use Disorder Symptoms at Age 35 in a National Sample

OBJECTIVE

The purpose of this longitudinal study was to identify associations of drinking intensity at age 29/30 with symptoms of alcohol use disorder (AUD) at age 35.

METHOD

Analyses used national longitudinal data from 1,253 individuals (53.5% female) participating in the Monitoring the Future study. Age 29/30 data were collected from 2005 to 2013; age 35 data were collected from 2010 to 2018. Multivariable models regressed age 35 past-5-year AUD symptoms (vs. nondisordered drinking/abstinence) on age 29/30 past-2-week drinking intensity (no/low [0-4] drinking, binge [5-9] drinking, high-intensity [10+] drinking), with key covariates being controlled for.

RESULTS

At age 35, 32.6% (SE = 1.50) of respondents reported AUD symptoms. AUD symptoms at age 35 were reported by 77.5% (SE = 4.79) of participants who reported age 29/30 high-intensity drinking and 60.6% (SE = 3.95) of participants who reported age 29/30 binge drinking. Age 35 past-5-year abstinence was reported by almost no respondents reporting age 29/30 binge drinking or high-intensity drinking. AUD symptoms at age 35 were significantly more likely for those who reported binge (adjusted multivariable odds ratio [AOR] = 5.61, 95% CI [3.79, 8.30], p < .001) or high-intensity (AOR = 12.26, 95% CI [6.70, 22.41], p < .001) drinking versus no/low drinking at age 29/30. The likelihood of having AUD symptoms was significantly higher for high-intensity than for binge drinkers (AOR = 2.18, 95% CI [1.14, 4.19], p = .019).

CONCLUSIONS

Nearly 80% of those young adults who reported engaging in high-intensity drinking (10+ drinks in a row) at age 29/30 later reported AUD symptoms at age 35. High-intensity drinking appears to be a strong prospective marker of risk for AUD symptoms among adults in the United States.

Drinking Intensity at Age 29/30 as a Predictor of Alcohol Use Disorder Symptoms at Age 35 in a National Sample

OBJECTIVE

The purpose of this longitudinal study was to identify associations of drinking intensity at age 29/30 with symptoms of alcohol use disorder (AUD) at age 35.

METHOD

Analyses used national longitudinal data from 1,253 individuals (53.5% female) participating in the Monitoring the Future study. Age 29/30 data were collected from 2005 to 2013; age 35 data were collected from 2010 to 2018. Multivariable models regressed age 35 past-5-year AUD symptoms (vs. nondisordered drinking/abstinence) on age 29/30 past-2-week drinking intensity (no/low [0-4] drinking, binge [5-9] drinking, high-intensity [10+] drinking), with key covariates being controlled for.

RESULTS

At age 35, 32.6% (SE = 1.50) of respondents reported AUD symptoms. AUD symptoms at age 35 were reported by 77.5% (SE = 4.79) of participants who reported age 29/30 high-intensity drinking and 60.6% (SE = 3.95) of participants who reported age 29/30 binge drinking. Age 35 past-5-year abstinence was reported by almost no respondents reporting age 29/30 binge drinking or high-intensity drinking. AUD symptoms at age 35 were significantly more likely for those who reported binge (adjusted multivariable odds ratio [AOR] = 5.61, 95% CI [3.79, 8.30], p < .001) or high-intensity (AOR = 12.26, 95% CI [6.70, 22.41], p < .001) drinking versus no/low drinking at age 29/30. The likelihood of having AUD symptoms was significantly higher for high-intensity than for binge drinkers (AOR = 2.18, 95% CI [1.14, 4.19], p = .019).

CONCLUSIONS

Nearly 80% of those young adults who reported engaging in high-intensity drinking (10+ drinks in a row) at age 29/30 later reported AUD symptoms at age 35. High-intensity drinking appears to be a strong prospective marker of risk for AUD symptoms among adults in the United States.

Causal inference with observational data: the need for triangulation of evidence

The goal of much observational research is to identify risk factors that have a causal effect on health and social outcomes. However, observational data are subject to biases from confounding, selection and measurement, which can result in an underestimate or overestimate of the effect of interest. Various advanced statistical approaches exist that offer certain advantages in terms of addressing these potential biases. However, although these statistical approaches have different underlying statistical assumptions, in practice they cannot always completely remove key sources of bias; therefore, using design-based approaches to improve causal inference is also important. Here it is the design of the study that addresses the problem of potential bias - either by ensuring it is not present (under certain assumptions) or by comparing results across methods with different sources and direction of potential bias. The distinction between statistical and design-based approaches is not an absolute one, but it provides a framework for triangulation - the thoughtful application of multiple approaches (e.g. statistical and design based), each with their own strengths and weaknesses, and in particular sources and directions of bias. It is unlikely that any single method can provide a definite answer to a causal question, but the triangulation of evidence provided by different approaches can provide a stronger basis for causal inference. Triangulation can be considered part of wider efforts to improve the transparency and robustness of scientific research, and the wider scientific infrastructure and system of incentives.

A Latent Transition Analysis of Self-Reported Reasons for Marijuana Use During Young Adulthood

Individuals' reasons for marijuana use have been linked to their risk for continued use and development of disordered use. Although individuals tend to have multiple reasons for use, co-occurrence of reasons is not always accounted for in analytic approaches. Latent transition analysis (LTA) is ideal for modeling transitions in co-occurring reasons. Using longitudinal panel data from Monitoring the Future, LTA was used to identify profiles of self-reported reasons for marijuana use among young adults, examine transitions between profiles, and determine whether cohort, gender, race/ethnicity, parent education, grade of first marijuana use, and 4-year college attendance predicted transitions between profiles. Data included senior year cohorts from 1976-2009 and were collected at ages 19/20, 21/22, and 23/24 (weighted n = 7,294; 55.9% female; 79.3% White). Five latent classes were identified: Non-Users and individuals with Experimental, Typical, Get High + Relax, and Escape + Coping Reasons. Transitions among Non-Users, Experimental Reasons, and Typical Reasons were common; generally, those with earlier cohort membership, early initiation, college non-attending parents, and college attendance were more likely to make transitions to higher-risk classes. As the legalization of recreational marijuana use continues to expand, change over time in reasons for use should be considered carefully as interventions are developed and implemented.

Motives for Alcohol and Marijuana Use as Predictors of Use and Problem Use Among Young Adult College Students

This study examined (a) differences between alcohol-only users and alcohol-marijuana co-users and (b) motives for use in relation to alcohol and marijuana use and problem use. Spring 2016 data among 1,870 past 4-month alcohol users (63.6% female, 69.1% White) from seven Georgia colleges/universities were analyzed cross-sectionally and with regard to problem use measured 4 months later. Correlates of co-use (n = 345; vs. alcohol-only use, n = 1,525) included greater alcohol and marijuana use frequency, problem drinking and marijuana use, and alcohol use motives (p's < .05). Controlling for covariates, alcohol use frequency correlated with greater marijuana use frequency and Coping and Self-enhancement alcohol use motives, but lower Conformity alcohol use motives (p's < .001); greater Coping and Self-enhancement alcohol use motives (p's < .01) predicted problem alcohol use. Marijuana use frequency correlated with greater Coping and Expansion marijuana use motives (p's < .05); greater Expansion marijuana use motives (p = .005) predicted problem marijuana use. College-based substance use interventions should target Coping and Self-enhancement alcohol use motives and Expansion marijuana use motives.

Advances in Statistical Methods for Causal Inference in Prevention Science: Introduction to the Special Section

The board of the Society for Prevention Research noted recently that extant methods for the analysis of causality mechanisms in prevention may still be too rudimentary for detailed and sophisticated analysis of causality hypotheses. This Special Section aims to fill some of the current voids, in particular in the domain of statistical methods of the analysis of causal inference. In the first article, Bray et al. propose a novel methodological approach in which they link propensity score techniques and Latent Class Analysis. In the second article, Kelcey et al. discuss power analysis tools for the study of causal mediation effects in cluster-randomized interventions. Wiedermann et al. present, in the third article, methods of Direction Dependence Analysis for the identification of confounders and for inference concerning the direction of causal effects in mediation models. A more general approach to the identification of causal structures in non-experimental data is presented by Shimizu in the fourth article. This approach is based on linear non-Gaussian acyclic models. Molenaar introduces vector-autoregressive methods for the optimal representation of Granger causality in time-dependent data. The Special Section concludes with a commentary by Musci and Stuart. In this commentary, the contributions of the articles in the Special Section are highlighted from the perspective of the experimental causal research tradition.

Sensitivity and specificity of information criteria

Information criteria (ICs) based on penalized likelihood, such as Akaike's information criterion (AIC), the Bayesian information criterion (BIC) and sample-size-adjusted versions of them, are widely used for model selection in health and biological research. However, different criteria sometimes support different models, leading to discussions about which is the most trustworthy. Some researchers and fields of study habitually use one or the other, often without a clearly stated justification. They may not realize that the criteria may disagree. Others try to compare models using multiple criteria but encounter ambiguity when different criteria lead to substantively different answers, leading to questions about which criterion is best. In this paper we present an alternative perspective on these criteria that can help in interpreting their practical implications. Specifically, in some cases the comparison of two models using ICs can be viewed as equivalent to a likelihood ratio test, with the different criteria representing different alpha levels and BIC being a more conservative test than AIC. This perspective may lead to insights about how to interpret the ICs in more complex situations. For example, AIC or BIC could be preferable, depending on the relative importance one assigns to sensitivity versus specificity. Understanding the differences and similarities among the ICs can make it easier to compare their results and to use them to make informed decisions.

Alcohol motivations and behaviors during months young adults experience social role transitions: Microtransitions in early adulthood

The study examines how young adult transitions into and out of social roles (i.e., starting or ending a job, a romantic relationship, school) are associated with drinking motives (coping, enhancement, social, and conformity) and alcohol use in a given month. A community sample of young adult drinkers (N = 767; 56.3% female; 59.3% White; Ages 18-23) completed 24 consecutive months of online surveys (N = 15,333 months of data) about the previous month's experiences, social role transitions, and alcohol use. During the 2-year data collection window, participants reported starting/ending a job (10.0%/8.2%), a relationship (2.7%/4.3%), and school (9.2%/17.4%). Between persons, those who more often started jobs were more likely to drink and those who more often ended jobs had higher enhancement motives; those who more often ended relationships were more likely to drink, have a greater number of drinks when drinking, and have higher coping and enhancement motives; and those who more often started relationships had higher conformity motives. Within persons, during months when a relationship ended, participants reported stronger coping motives, and during months when a relationship started, they reported stronger social motives for drinking. During months when a relationship started or ended, participants also reported consuming a greater number of drinks when drinking. There were no differences based on starting or ending school. Young adult social role transitions are associated with concurrent changes in both alcohol use and motives for drinking. Understanding these contextual changes and their concomitant risks is key to providing salient interventions to reduce alcohol-related harm. (PsycINFO Database Record (c) 2018 APA, all rights reserved).

Ensuring Causal, Not Casual, Inference

With innovation in causal inference methods and a rise in non-experimental data availability, a growing number of prevention researchers and advocates are thinking about causal inference. In this commentary, we discuss the current state of science as it relates to causal inference in prevention research, and reflect on key assumptions of these methods. We review challenges associated with the use of causal inference methodology, as well as considerations for hoping to integrate causal inference methods into their research. In short, this commentary addresses the key concepts of causal inference and suggests a greater emphasis on thoughtfully designed studies (to avoid the need for strong and potentially untestable assumptions) combined with analyses of sensitivity to those assumptions.