New variable selection methods for zero-inflated count data with applications to the substance abuse field

Identifying high-risk areas for dog-mediated rabies using Bayesian spatial regression

Despite ongoing control efforts, rabies remains an endemic zoonotic disease in many countries. Determining high-risk areas and the space-time patterns of rabies spread, as it relates to epidemiologically important factors, can support policymakers and program managers alike to develop evidence-based targeted surveillance and control programs. In this One Health approach which selected Thailand as the example site, the location-based risk of contracting dog-mediated rabies by both human and animal populations was quantified using a Bayesian spatial regression model. Specifically, a conditional autoregressive (CAR) Bayesian zero-inflated Poisson (ZIP) regression was fitted to the reported human and animal rabies case counts of each district, from the 2012-2017 period. The human population was used as an offset. The epidemiologically important factors hypothesized as risk modifiers and therefore tested as predictors included: number of dog bites/attacks, the population of dogs and cats, number of Buddhist temples, garbage dumps, animal vaccination, post-exposure prophylaxis, poverty, and shared administrative borders. Disparate sources of data were used to improve the estimated associations and predictions. Model performance was assessed using cross-validation. Results suggested that accounting for the association between human and animal rabies with number of dog bites/attacks, number of owned and un-owned dogs; shared country borders, number of Buddhist temples, poverty levels, and accounting for spatial dependence between districts, may help to predict the risk districts for dog-mediated rabies in Thailand. The fitted values of the spatial regression were mapped to illustrate the risk of dog-mediated rabies. The cross-validation indicated an adequate performance of the spatial regression model (AUC = 0.81), suggesting that had this spatial regression approach been used to identify districts at risk in 2015, the cases reported in 2016/17 would have been predicted with model sensitivity and specificity of 0.71 and 0.80, respectively. While active surveillance is ideal, this approach of using multiple data sources to improve risk estimation may inform current rabies surveillance and control efforts including determining rabies-free zones, and the roll-out of human post-exposure prophylaxis and anti-rabies vaccines for animals in determining high-risk areas.

Investigating factors associated with the number of rehospitalizations among patients with schizophrenia disorder using penalized count regression models

BACKGROUND

Schizophrenia is a chronic, severe, and debilitating mental disorder always considered one of the recurrent psychiatric diseases. This study aimed to use penalized count regression models to determine factors associated with the number of rehospitalizations of schizophrenia disorder.

METHODS

This retrospective cohort study was performed on 413 schizophrenic patients who had been referred to the Sina (Farshchian) Educational and Medical Center in Hamadan, Iran, between March 2011 and March 2019. The penalized count regression models were fitted using R.3.5.2.

RESULTS

About 73% of the patients were male. The mean (SD) of age and the number of rehospitalizations were 36.16 (11.18) years and 1.21 (2.18), respectively. According to the results, longer duration of illness (P < 0.001), having a positive family history of psychiatric illness (P = 0.017), having at least three children (P = 0.013), unemployment, disability, and retirement (P = 0.025), residence in other Hamadan province townships (P = 0.003) and having a history of arrest/prison (P = 0.022) were significantly associated with an increase in the number of rehospitalizations.

CONCLUSION

To reduce the number of rehospitalizations among schizophrenic patients, it is recommended to provide special medical services for patients who do not have access to specialized medical centers and to create the necessary infrastructure for the employment of patients.

Epidemiological characterization of Clonorchis sinensis infection in humans and freshwater fish in Guangxi, China

Background

Clonorchiasis is a widespread yet neglected foodborne disease with over 85% of all cases found in China. Guangxi province, located in southeastern China, ranks among the highest endemic provinces. We explore the epidemiological status and determinants of Clonorchis sinensis (C. sinensis) infection in humans and freshwater fish in Guangxi, China.

Methods

Data on C. sinensis infection in humans from January 2008 to December 2017were extracted from the China Information System for Disease Control and Prevention. An active surveillance of C. sinensis infection in fish was conducted in 2016–2017. County level data including potential environmental, social-economical and behavioral determinants was also collected. Univariate and multivariate logistic regression models were used to explore the determinants of C. sinensis infection in humans and fish. Simple and multiple zero-inflated Poisson regression models were fit to assess the associated factors of clonorchiasis in humans at the county level.

Results

Totally, 4526 C. sinensis cases were reported between 2008 and 2017, with an annual prevalencerate of 0.96/100,000 persons. Of 101 counties in Guangxi, 97 reported at least 1 case. Among 2,098 fish samples, 203 (9.7%) from 70 counties contained C. sinensis. The rate was higher in small fish including Pseudorasbora parva (45.3%), Misgurnus anguillicaudatus (41.2%), Hemicculter leuciclus (34.5%), unclassified small fishes (30.9%), Cyprinidae (20.0%), Cirrhinus molitorella (16.4%), Carassius auratus (13.6%) and Cyprinus carpio (13.3%), while it was lower in fish species that are usually used in preparing raw fish dishes including Ctenopharyngodon idellus (3.6%), Spinibarbus denticulatus (3.7%), Monopterus albus (6.4%), Cyprinus carpio (4.4%), Oreochromis mossambicus (3.3%) and Spualiobarbus Curriculus (6.6%). The C. sinensis infection in fish was only associated with fish species. The estimated human clonorchiasis prevalence at the county level was positively associated with raw fish consumption habits and certain rivers.

Conclusions

Clonorchiasis is highly prevalent in both humans and freshwater fish in Guangxi. Environmental, social-economic and behavioral determinants contribute to the high prevalence as well as the significant differential distribution by county. Regular surveillance should be implemented for clonorchiasis to demonstrate the change in epidemiology and burden, which will benefit the design of interventions.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-022-07244-2.

Variable selection for longitudinal zero-inflated power series transition model

In modeling many longitudinal count clinical studies, the excess of zeros is a common problem. To take into account the extra zeros, the zero-inflated power series (ZIPS) models have been applied. These models assume a latent mixture model consisting of a count component and a degenerated zero component that has a unit point mass at zero. Usually, the current response measurement in a longitudinal sequence is a function of previous outcomes. For example, in a study about acute renal allograft rejection, the number of acute rejection episodes for a patient in current time is a function of this outcome at previous follow-up times. In this paper, we consider a transition model for accounting the dependence of current outcome on the previous outcomes in the presence of excess zeros. New variable selection methods for the ZIPS transition model using least absolute shrinkage and selection operator (LASSO), minimax concave penalty (MCP) and smoothly clipped absolute deviation (SCAD) penalties are proposed. An expectation-maximization (EM) algorithm using the penalized likelihood is applied for both parameters estimations and conducting variable selection. Some simulation studies are performed to investigate the performance of the proposed approach and the approach is applied to analyze a real dataset.

Identifying factors associated with opioid cessation in a biracial sample using machine learning

AIM

Racial disparities in opioid use disorder (OUD) management exist, however, and there is limited research on factors that influence opioid cessation in different population groups.

METHODS

We employed multiple machine learning prediction algorithms least absolute shrinkage and selection operator, random forest, deep neural network, and support vector machine to assess factors associated with ceasing opioid use in a sample of 1,192 African Americans (AAs) and 2,557 individuals of European ancestry (EAs) who met Diagnostic and Statistical Manual of Mental Disorders, 5th Edition criteria for OUD. Values for nearly 4,000 variables reflecting demographics, alcohol and other drug use, general health, non-drug use behaviors, and diagnoses for other psychiatric disorders, were obtained for each participant from the Semi-Structured Assessment for Drug Dependence and Alcoholism, a detailed semi-structured interview.

RESULTS

Support vector machine models performed marginally better on average than other machine learning methods with maximum prediction accuracies of 75.4% in AAs and 79.4% in EAs. Subsequent stepwise regression considered the 83 most highly ranked variables across all methods and models and identified less recent cocaine use (AAs: odds ratio (OR) = 1.82, P = 9.19 × 10-5; EAs: OR = 1.91, P = 3.30 × 10-15), shorter duration of opioid use (AAs: OR = 0.55, P = 5.78 × 10-6; EAs: OR = 0.69, P = 3.01 × 10-7), and older age (AAs: OR = 2.44, P = 1.41 × 10-12; EAs: OR = 2.00, P = 5.74 × 10-9) as the strongest independent predictors of opioid cessation in both AAs and EAs. Attending self-help groups for OUD was also an independent predictor (P < 0.05) in both population groups, while less gambling severity (OR = 0.80, P = 3.32 × 10-2) was specific to AAs and post-traumatic stress disorder recovery (OR = 1.93, P = 7.88 × 10-5), recent antisocial behaviors (OR = 0.64, P = 2.69 × 10-3), and atheism (OR = 1.45, P = 1.34 × 10-2) were specific to EAs. Factors related to drug use comprised about half of the significant independent predictors in both AAs and EAs, with other predictors related to non-drug use behaviors, psychiatric disorders, overall health, and demographics.

CONCLUSIONS

These proof-of-concept findings provide avenues for hypothesis-driven analysis, and will lead to further research on strategies to improve OUD management in EAs and AAs.

Mediation analysis with zero-inflated substance use outcomes: Challenges and recommendations

Mediating mechanisms are important components of substance use research, as many substance use interventions work by targeting mediating variables. One issue that is common in substance use research is the presence of many responses of zero in a count variable that is the primary outcome of interest, such as number of drinks per week or number of substances used in the past month. The goal of this paper is to highlight the unique challenges that substance use researchers face when conducting mediation analysis with a zero-inflated count outcome. In this paper, we first describe the models that are commonly used for zero-inflated count data, and when it is appropriate to use them. We then describe general mediation analysis and summarize the small body of work that has focused on mediation for count and zero-inflated count outcomes. We identify the main issue of computing the mediated effect when outcomes are zero-inflated, namely, that the path leading to the zero-inflated count outcome (or mediator) is modeled in two parts. We then provide two examples of mediation models with different conclusions that have zero-inflated count outcomes using adolescent substance use data and define the issues that arise when assessing mediation for each. Finally, we describe the directions in which we must target future methodological research to create accessible solutions for handling mediation with zero-inflated count data in substance use research.

A GEE-type approach to untangle structural and random zeros in predictors

Count outcomes with excessive zeros are common in behavioral and social studies, and zero-inflated count models such as zero-inflated Poisson (ZIP) and zero-inflated Negative Binomial (ZINB) can be applied when such zero-inflated count data are used as response variable. However, when the zero-inflated count data are used as predictors, ignoring the difference of structural and random zeros can result in biased estimates. In this paper, a generalized estimating equation (GEE)-type mixture model is proposed to jointly model the response of interest and the zero-inflated count predictors. Simulation studies show that the proposed method performs well for practical settings and is more robust for model misspecification than the likelihood-based approach. A case study is also provided for illustration.

A Longitudinal Examination of Decisions to Ride and Decline Rides with Drinking Drivers

BACKGROUND

Riding with a drinking driver (RWDD) is a serious concern that leads to numerous preventable deaths every year. There is a significant gap in research on empirically tested predictors of RWDD that could be implemented in prevention efforts. College students are in need of such prevention efforts, as they have some of the highest rates of alcohol-related crash fatalities and may engage in RWDD more than their noncollege peers. This study utilized behavioral decision-making approach to examine predictors of RWDD and declining a ride from a drinking driver (Decline) in older college students.

METHODS

Students (n = 791) in their third year of college were enrolled from 3 large and diverse universities. Psychosocial (e.g., expectancies, norms) and decision-making variables (willingness to RWDD and intentions to use alternatives) were assessed in the fall of their third year. One year later, RWDD and Decline behaviors were assessed. Zero-inflated Poisson analyses were used to assess how decision-making variables predicted RWDD and Decline behavior. Associations between psychosocial and decision-making variables were also assessed.

RESULTS

Thirteen percent of students reported RWDD and ~28% reported Decline behavior. Willingness to RWDD and typical weekly drinking were both associated with increases in RWDD (odds ratio [OR] = 1.58 and 1.40, respectively), whereas intentions to use alternatives, sex, and ethnicity were not associated with RWDD. Only weekly drinking was associated with Decline, with an increase in drinking associated with increased Decline (OR = 1.48). All psychosocial variables were significantly associated with the decision-making variables except positive expectancies.

CONCLUSIONS

Results provide evidence that willingness to RWDD is a predictor of future RWDD, even if students intend to use safe alternatives. Future research is needed to better understand decision-making factors that influence Decline. Results also suggest prevention and interventions efforts, such as brief motivational intervention, Parent-Based Interventions, and normative feedback interventions could be adapted to reduce RWDD.

Group regularization for zero-inflated negative binomial regression models with an application to health care demand in Germany

In many biomedical applications, covariates are naturally grouped, with variables in the same group being systematically related or statistically correlated. Under such settings, variable selection must be conducted at both group and individual variable levels. Motivated by the widespread availability of zero-inflated count outcomes and grouped covariates in many practical applications, we consider group regularization for zero-inflated negative binomial regression models. Using a least squares approximation of the mixture likelihood and a variety of group-wise penalties on the coefficients, we propose a unified algorithm (Gooogle: Group Regularization for Zero-inflated Count Regression Models) to efficiently compute the entire regularization path of the estimators. We investigate the finite sample performance of these methods through extensive simulation experiments and the analysis of a German health care demand dataset. Finally, we derive theoretical properties of these methods under reasonable assumptions, which further provides deeper insight into the asymptotic behavior of these approaches. The open source software implementation of this method is publicly available at: https://github.com/himelmallick/Gooogle.

Reasons for Testing Mediation in the Absence of an Intervention Effect: A Research Imperative in Prevention and Intervention Research

OBJECTIVE

Mediation models are used in prevention and intervention research to assess the mechanisms by which interventions influence outcomes. However, researchers may not investigate mediators in the absence of intervention effects on the primary outcome variable. There is emerging evidence that in some situations, tests of mediated effects can be statistically significant when the total intervention effect is not statistically significant. In addition, there are important conceptual and practical reasons for investigating mediation when the intervention effect is nonsignificant.

METHOD

This article discusses the conditions under which mediation may be present when an intervention effect does not have a statistically significant effect and why mediation should always be considered important.

RESULTS

Mediation may be present in the following conditions: when the total and mediated effects are equal in value, when the mediated and direct effects have opposing signs, when mediated effects are equal across single and multiple-mediator models, and when specific mediated effects have opposing signs. Mediation should be conducted in every study because it provides the opportunity to test known and replicable mediators, to use mediators as an intervention manipulation check, and to address action and conceptual theory in intervention models.

CONCLUSIONS

Mediators are central to intervention programs, and mediators should be investigated for the valuable information they provide about the success or failure of interventions.

Network analysis for count data with excess zeros

BACKGROUND

Undirected graphical models or Markov random fields have been a popular class of models for representing conditional dependence relationships between nodes. In particular, Markov networks help us to understand complex interactions between genes in biological processes of a cell. Local Poisson models seem to be promising in modeling positive as well as negative dependencies for count data. Furthermore, when zero counts are more frequent than are expected, excess zeros should be considered in the model.

METHODS

We present a penalized Poisson graphical model for zero inflated count data and derive an expectation-maximization (EM) algorithm built on coordinate descent. Our method is shown to be effective through simulated and real data analysis.

RESULTS

Results from the simulated data indicate that our method outperforms the local Poisson graphical model in the presence of excess zeros. In an application to a RNA sequencing data, we also investigate the gender effect by comparing the estimated networks according to different genders. Our method may help us in identifying biological pathways linked to sex hormone regulation and thus understanding underlying mechanisms of the gender differences.

CONCLUSIONS

We have presented a penalized version of zero inflated spatial Poisson regression and derive an efficient EM algorithm built on coordinate descent. We discuss possible improvements of our method as well as potential research directions associated with our findings from the RNA sequencing data.

Untangle the Structural and Random Zeros in Statistical Modelings

Count data with structural zeros are common in public health applications. There are considerable researches focusing on zero-inflated models such as zero-inflated Poisson (ZIP) and zero-inflated Negative Binomial (ZINB) models for such zero-inflated count data when used as response variable. However, when such variables are used as predictors, the difference between structural and random zeros is often ignored and may result in biased estimates. One remedy is to include an indicator of the structural zero in the model as a predictor if observed. However, structural zeros are often not observed in practice, in which case no statistical method is available to address the bias issue. This paper is aimed to fill this methodological gap by developing parametric methods to model zero-inflated count data when used as predictors based on the maximum likelihood approach. The response variable can be any type of data including continuous, binary, count or even zero-inflated count responses. Simulation studies are performed to assess the numerical performance of this new approach when sample size is small to moderate. A real data example is also used to demonstrate the application of this method.

Stochastic variable selection strategies for zero-inflated models

When count data exhibit excess zero, that is more zero counts than a simpler parametric distribution can model, the zero-inflated Poisson (ZIP) or zero-inflated negative binomial (ZINB) models are often used. Variable selection for these models is even more challenging than for other regression situations because the availability of p covariates implies 4 p possible models. We adapt to zero-inflated models an approach for variable selection that avoids the screening of all possible models. This approach is based on a stochastic search through the space of all possible models, which generates a chain of interesting models. As an additional novelty, we propose three ways of extracting information from this rich chain and we compare them in two simulation studies, where we also contrast our approach with regularization (penalized) techniques available in the literature. The analysis of a typical dataset that has motivated our research is also presented, before concluding with some recommendations.

Chaos to complexity: leveling the playing field for measuring value in primary care

RATIONALE, AIMS AND OBJECTIVES

Develop a risk-stratification model that clusters primary care patients with similar co-morbidities and social determinants and ranks 'within-practice' clusters of complex patients based on likelihood of hospital and emergency department (ED) utilization.

METHODS

A retrospective cohort analysis was performed on 10 408 adults who received their primary care at the Medical University of South Carolina University Internal Medicine clinic. A two-part generalized linear regression model was used to fit a predictive model for ED and hospital utilization. Agglomerative hierarchical clustering was used to identify patient subgroups with similar co-morbidities.

RESULTS

Factors associated with increased risk of utilization included specific disease clusters {e.g. renal disease cluster [rate ratio, RR = 5.47; 95% confidence interval (CI; 4.54, 6.59) P < 0.0001]}, low clinic visit adherence [RR = 0.33; 95% CI (0.28, 0.39) P < 0.0001] and census measure of high poverty rate [RR = 1.20; 95% CI (1.11, 1.28) P < 0.0001]. In the cluster model, a stable group of four clusters remained regardless of the number of additional clusters forced into the model. Although the largest number of high-utilization patients (top 20%) was in the multiple chronic condition cluster (1110 out of 4728), the largest proportion of high-utilization patients was in the renal disease cluster (67%).

CONCLUSIONS

Risk stratification enhanced with disease clustering organizes a primary care population into groups of similarly complex patients so that care coordination efforts can be focused and value of care can be maximized.

A time-varying effect model for examining group differences in trajectories of zero-inflated count outcomes with applications in substance abuse research

This study proposes a time-varying effect model for examining group differences in trajectories of zero-inflated count outcomes. The motivating example demonstrates that this zero-inflated Poisson model allows investigators to study group differences in different aspects of substance use (e.g., the probability of abstinence and the quantity of alcohol use) simultaneously. The simulation study shows that the accuracy of estimation of trajectory functions improves as the sample size increases; the accuracy under equal group sizes is only higher when the sample size is small (100). In terms of the performance of the hypothesis testing, the type I error rates are close to their corresponding significance levels under all settings. Furthermore, the power increases as the alternative hypothesis deviates more from the null hypothesis, and the rate of this increasing trend is higher when the sample size is larger. Moreover, the hypothesis test for the group difference in the zero component tends to be less powerful than the test for the group difference in the Poisson component. Copyright © 2016 John Wiley & Sons, Ltd.

Identifying Pleiotropic Genes in Genome-Wide Association Studies for Multivariate Phenotypes with Mixed Measurement Scales

We propose a multivariate genome-wide association test for mixed continuous, binary, and ordinal phenotypes. A latent response model is used to estimate the correlation between phenotypes with different measurement scales so that the empirical distribution of the Fisher’s combination statistic under the null hypothesis is estimated efficiently. The simulation study shows that our proposed correlation estimation methods have high levels of accuracy. More importantly, our approach conservatively estimates the variance of the test statistic so that the type I error rate is controlled. The simulation also shows that the proposed test maintains the power at the level very close to that of the ideal analysis based on known latent phenotypes while controlling the type I error. In contrast, conventional approaches–dichotomizing all observed phenotypes or treating them as continuous variables–could either reduce the power or employ a linear regression model unfit for the data. Furthermore, the statistical analysis on the database of the Study of Addiction: Genetics and Environment (SAGE) demonstrates that conducting a multivariate test on multiple phenotypes can increase the power of identifying markers that may not be, otherwise, chosen using marginal tests. The proposed method also offers a new approach to analyzing the Fagerström Test for Nicotine Dependence as multivariate phenotypes in genome-wide association studies.

Two-stage model for time varying effects of zero-inflated count longitudinal covariates with applications in health behaviour research

This study proposes a two-stage approach to characterize individual developmental trajectories of health risk behaviors and delineate their time-varying effects on short-term or long-term health outcomes. Our model can accommodate longitudinal covariates with zero-inflated counts and discrete outcomes. The longitudinal data of a well-known study of youth at high risk for substance abuse are presented as a motivating example to demonstrate the effectiveness of the model in delineating critical developmental periods of prevention and intervention. Our simulation study shows that the performance of the proposed model improves as the sample size or number of time points increases. When there are excess zeros in the data, the regular Poisson model cannot estimate either the longitudinal covariate process or its time-varying effect well. This result, therefore, emphasizes the important role that the proposed model plays in handling zero-inflation in the data.

EM Adaptive LASSO-A Multilocus Modeling Strategy for Detecting SNPs Associated with Zero-inflated Count Phenotypes

Count data are increasingly ubiquitous in genetic association studies, where it is possible to observe excess zero counts as compared to what is expected based on standard assumptions. For instance, in rheumatology, data are usually collected in multiple joints within a person or multiple sub-regions of a joint, and it is not uncommon that the phenotypes contain enormous number of zeroes due to the presence of excessive zero counts in majority of patients. Most existing statistical methods assume that the count phenotypes follow one of these four distributions with appropriate dispersion-handling mechanisms: Poisson, Zero-inflated Poisson (ZIP), Negative Binomial, and Zero-inflated Negative Binomial (ZINB). However, little is known about their implications in genetic association studies. Also, there is a relative paucity of literature on their usefulness with respect to model misspecification and variable selection. In this article, we have investigated the performance of several state-of-the-art approaches for handling zero-inflated count data along with a novel penalized regression approach with an adaptive LASSO penalty, by simulating data under a variety of disease models and linkage disequilibrium patterns. By taking into account data-adaptive weights in the estimation procedure, the proposed method provides greater flexibility in multi-SNP modeling of zero-inflated count phenotypes. A fast coordinate descent algorithm nested within an EM (expectation-maximization) algorithm is implemented for estimating the model parameters and conducting variable selection simultaneously. Results show that the proposed method has optimal performance in the presence of multicollinearity, as measured by both prediction accuracy and empirical power, which is especially apparent as the sample size increases. Moreover, the Type I error rates become more or less uncontrollable for the competing methods when a model is misspecified, a phenomenon routinely encountered in practice.

Variable selection for distribution-free models for longitudinal zero-inflated count responses

Zero-inflated count outcomes arise quite often in research and practice. Parametric models such as the zero-inflated Poisson and zero-inflated negative binomial are widely used to model such responses. Like most parametric models, they are quite sensitive to departures from assumed distributions. Recently, new approaches have been proposed to provide distribution-free, or semi-parametric, alternatives. These methods extend the generalized estimating equations to provide robust inference for population mixtures defined by zero-inflated count outcomes. In this paper, we propose methods to extend smoothly clipped absolute deviation (SCAD)-based variable selection methods to these new models. Variable selection has been gaining popularity in modern clinical research studies, as determining differential treatment effects of interventions for different subgroups has become the norm, rather the exception, in the era of patent-centered outcome research. Such moderation analysis in general creates many explanatory variables in regression analysis, and the advantages of SCAD-based methods over their traditional counterparts render them a great choice for addressing this important and timely issues in clinical research. We illustrate the proposed approach with both simulated and real study data. Copyright © 2016 John Wiley & Sons, Ltd.

A time-varying effect model for studying gender differences in health behavior

This study proposes a time-varying effect model that can be used to characterize gender-specific trajectories of health behaviors and conduct hypothesis testing for gender differences. The motivating examples demonstrate that the proposed model is applicable to not only multi-wave longitudinal studies but also short-term studies that involve intensive data collection. The simulation study shows that the accuracy of estimation of trajectory functions improves as the sample size and the number of time points increase. In terms of the performance of the hypothesis testing, the type I error rates are close to their corresponding significance levels under all combinations of sample size and number of time points. Furthermore, the power increases as the alternative hypothesis deviates more from the null hypothesis, and the rate of this increasing trend is higher when the sample size and the number of time points are larger.

Current intimate relationship status, depression, and alcohol use among bisexual women: The mediating roles of bisexual-specific minority stressors

Current intimate relationship characteristics, including gender and number of partner(s), may affect one's visibility as a bisexual individual and the minority stressors they experience, which may in turn influence their health. The current study tested four hypotheses: 1) minority stressors vary by current intimate relationship status; 2) higher minority stressors are associated with higher depressive symptoms and alcohol-related outcomes; 3) depressive symptoms and alcohol-related outcomes vary by current intimate relationship status; and 4) minority stressors will mediate differences in these outcomes. Participants included 470 self-identified bisexual women (65% Caucasian, mean age: 21) from a sample of sexual minority women recruited from different geographic regions in the United States through advertisements on social networking sites and Craigslist. Participants completed a 45 minute survey. Respondents with single partners were first grouped by partner gender (male partner: n=282; female partner: n=56). Second, women were grouped by partner gender/number (single female/male partner: n = 338; women with multiple female and male partners: n=132). Women with single male partners and women with multiple male and female partners exhibited elevated experienced bi-negativity and differences in outness (H1). Experienced and internalized bi-negativity were associated with health outcomes, but not outness (H2). Differences in outcomes emerged by partner number and partner number/gender (H3); these differences were mediated by experienced bi-negativity (H4). These results suggest that experiences of discrimination may underlie differences in health related to bisexual women's relationship structure and highlight the importance of evaluating women's relational context as well as sexual identification in understanding health risk behaviors.

Variable selection for zero-inflated and overdispersed data with application to health care demand in Germany

In health services and outcome research, count outcomes are frequently encountered and often have a large proportion of zeros. The zero-inflated negative binomial (ZINB) regression model has important applications for this type of data. With many possible candidate risk factors, this paper proposes new variable selection methods for the ZINB model. We consider maximum likelihood function plus a penalty including the least absolute shrinkage and selection operator (LASSO), smoothly clipped absolute deviation (SCAD), and minimax concave penalty (MCP). An EM (expectation-maximization) algorithm is proposed for estimating the model parameters and conducting variable selection simultaneously. This algorithm consists of estimating penalized weighted negative binomial models and penalized logistic models via the coordinated descent algorithm. Furthermore, statistical properties including the standard error formulae are provided. A simulation study shows that the new algorithm not only has more accurate or at least comparable estimation, but also is more robust than the traditional stepwise variable selection. The proposed methods are applied to analyze the health care demand in Germany using the open-source R package mpath.

Functional linear models for zero-inflated count data with application to modeling hospitalizations in patients on dialysis

We propose functional linear models for zero-inflated count data with a focus on the functional hurdle and functional zero-inflated Poisson (ZIP) models. Although the hurdle model assumes the counts come from a mixture of a degenerate distribution at zero and a zero-truncated Poisson distribution, the ZIP model considers a mixture of a degenerate distribution at zero and a standard Poisson distribution. We extend the generalized functional linear model framework with a functional predictor and multiple cross-sectional predictors to model counts generated by a mixture distribution. We propose an estimation procedure for functional hurdle and ZIP models, called penalized reconstruction, geared towards error-prone and sparsely observed longitudinal functional predictors. The approach relies on dimension reduction and pooling of information across subjects involving basis expansions and penalized maximum likelihood techniques. The developed functional hurdle model is applied to modeling hospitalizations within the first 2 years from initiation of dialysis, with a high percentage of zeros, in the Comprehensive Dialysis Study participants. Hospitalization counts are modeled as a function of sparse longitudinal measurements of serum albumin concentrations, patient demographics, and comorbidities. Simulation studies are used to study finite sample properties of the proposed method and include comparisons with an adaptation of standard principal components regression.

Structural zeroes and zero-inflated models

Summary

In psychosocial and behavioral studies count outcomes recording the frequencies of the occurrence of some health or behavior outcomes (such as the number of unprotected sexual behaviors during a period of time) often contain a preponderance of zeroes because of the presence of ‘structural zeroes’ that occur when some subjects are not at risk for the behavior of interest. Unlike random zeroes (responses that can be greater than zero, but are zero due to sampling variability), structural zeroes are usually very different, both statistically and clinically. False interpretations of results and study findings may result if differences in the two types of zeroes are ignored. However, in practice, the status of the structural zeroes is often not observed and this latent nature complicates the data analysis. In this article, we focus on one model, the zero-inflated Poisson (ZIP) regression model that is commonly used to address zero-inflated data. We first give a brief overview of the issues of structural zeroes and the ZIP model. We then given an illustration of ZIP with data from a study on HIV-risk sexual behaviors among adolescent girls. Sample codes in SAS and Stata are also included to help perform and explain ZIP analyses.

EM for regularized zero-inflated regression models with applications to postoperative morbidity after cardiac surgery in children

This paper proposes a new statistical approach for predicting postoperative morbidity such as intensive care unit length of stay and number of complications after cardiac surgery in children. In a recent multi-center study sponsored by the National Institutes of Health, 311 children undergoing cardiac surgery were enrolled. Morbidity data are count data in which the observations take only nonnegative integer values. Often, the number of zeros in the sample cannot be accommodated properly by a simple model, thus requiring a more complex model such as the zero-inflated Poisson regression model. We are interested in identifying important risk factors for postoperative morbidity among many candidate predictors. There is only limited methodological work on variable selection for the zero-inflated regression models. In this paper, we consider regularized zero-inflated Poisson models through penalized likelihood function and develop a new expectation-maximization algorithm for numerical optimization. Simulation studies show that the proposed method has better performance than some competing methods. Using the proposed methods, we analyzed the postoperative morbidity, which improved the model fitting and identified important clinical and biomarker risk factors.

Changes in substance use-related health risk behaviors on the timeline follow-back interview as a function of length of recall period

The timeline follow-back (TLFB) interview was adopted to collect retrospective data on daily substance use and violence from 598 youth seeking care in an urban Emergency Department in Flint, Michigan during 2009-2011. Generalized linear mixed models with flexible smooth functions of time were employed to characterize the change in risk behaviors as a function of the length of recall period. Our results suggest that the 1-week recall period may be more effective for capturing atypical or variable patterns of risk behaviors, whereas a recall period longer than 2 weeks may result in a more stable estimation of a typical pattern.

On the implication of structural zeros as independent variables in regression analysis: applications to alcohol research

In alcohol studies, drinking outcomes such as number of days of any alcohol drinking (DAD) over a period of time do not precisely capture the differences among subjects in a study population of interest. For example, the value of 0 on DAD could mean that the subject was continually abstinent from drinking such as lifetime abstainers or the subject was alcoholic, but happened not to use any alcohol during the period of interest. In statistics, zeros of the first kind are called structural zeros, to distinguish them from the sampling zeros of the second type. As the example indicates, the structural and sampling zeros represent two groups of subjects with quite different psychosocial outcomes. In the literature on alcohol use, although many recent studies have begun to explicitly account for the differences between the two types of zeros in modeling drinking variables as a response, none has acknowledged the implications of the different types of zeros when such modeling drinking variables are used as a predictor. This paper serves as the first attempt to tackle the latter issue and illustrate the importance of disentangling the structural and sampling zeros by using simulated as well as real study data.

Risk factor selection in rate making: EM adaptive LASSO for zero-inflated poisson regression models

Risk factor selection is very important in the insurance industry, which helps precise rate making and studying the features of high-quality insureds. Zero-inflated data are common in insurance, such as the claim frequency data, and zero-inflation makes the selection of risk factors quite difficult. In this article, we propose a new risk factor selection approach, EM adaptive LASSO, for a zero-inflated Poisson regression model, which combines the EM algorithm and adaptive LASSO penalty. Under some regularity conditions, we show that, with probability approaching 1, important factors are selected and the redundant factors are excluded. We investigate the finite sample performance of the proposed method through a simulation study and the analysis of car insurance data from SAS Enterprise Miner database.

Statistical models for longitudinal zero-inflated count data with applications to the substance abuse field

This study fills in the current knowledge gaps in statistical analysis of longitudinal zero-inflated count data by providing a comprehensive review and comparison of the hurdle and zero-inflated Poisson models in terms of the conceptual framework, computational advantage, and performance under different real data situations. The design of simulations represents the special features of a well-known longitudinal study of alcoholism so that the results can be generalizable to the substance abuse field. When the hurdle model is more natural under the conceptual framework of the data, the zero-inflated Poisson model tends to produce inaccurate estimates. Model performance improves with larger sample sizes, lower proportions of missing data, and lower correlations between covariates. The simulation also shows that the computational strength of the hurdle model disappears when random effects are included.

Early onset problem behaviors and alcohol, tobacco, and other substance use disorders in young adulthood

OBJECTIVE

Ten early onset problem behaviors were used to prospectively predict alcohol, tobacco, cannabis, and cocaine disorders in young adulthood (mean age=28.6 yrs) for a U.S. community sample of 671 participants.

METHOD

Data from a longitudinal study of participants who were recruited from high schools during adolescence and followed into young adulthood were used to evaluate prospective associations. The relationship between early onset problem behaviors, reported when participants were age 16 years, and psychiatric diagnoses assessed in young adulthood was tested. Structural equation models were used to evaluate both generality and specificity hypotheses regarding relationships between early onset problem behaviors and young adult disorders.

RESULTS

Findings supported the specificity hypothesis in that "like" early onset problem behaviors significantly predicted "like" young adult outcomes (e.g., early cocaine use predicted cocaine disorders). Furthermore, eliminating such "like" predictors in regression equations resulted in a 36% reduction in the amount of variance accounted for by the equation. The generality hypothesis was also supported in that a larger number of early onset problem behaviors strengthened the prediction of young adult disorders beyond the "like" attribute, and a dose-response pattern indicated that additional early onset problem behaviors increased the probable occurrence of a young adult disorder.

CONCLUSIONS

A comprehensive framework relating early onset problem behaviors to young adult substance disorders will require the integration of both generality and specificity hypotheses, and a developmental orientation focused on the unfolding of mediating and moderating processes. Early screening of multiple, rather than single, early onset problems is also discussed.