Modeling longitudinal binomial responses: implications from two dueling paradigms

Long-Term Socioeconomic and Mental Health Changes After Out-of-Hospital Cardiac Arrest in Women and Men

BACKGROUND

Long-term effects of out-of-hospital cardiac arrest (OHCA) may affect the ability to work and mental health. Our aim was to analyze 5-year changes in socioeconomic and mental health outcomes after OHCA in women and men.

METHODS

We included 259 women and 996 men from North Holland, the Netherlands, who survived 30 days after OHCA occurred between 2009 and 2015. We assessed changes in employment, income, primary earner status, and anxiety/depression (using medication proxies) from the year before the OHCA to 5 years after with generalized linear mixed models, stratified by sex. We tested differences in changes by sex with interaction terms. Additionally, we explored yearly changes. The 5-year changes after OHCA were compared with changes in a sex- and age-matched sample of people without OHCA. Differences were tested using an interaction term of time and OHCA status.

RESULTS

In both women and men (median age [Q1, Q3]: 51 [45, 55] and 54 [48, 57] years, respectively), decreases from before OHCA to 5 years thereafter were observed in the proportion employed (from 72.8% to 53.4% [women] and 80.9% to 63.7% [men]) and the median income. No change in primary earner status was observed in either sex. Dispensing of anxiety/depression medication increased only in women, especially after 1 year (odds ratio, 5.68 [95% CI, 2.05-15.74]) and 5 years (odds ratio, 5.73 [95% CI, 1.88-17.53]). Notable differences between women and men were observed for changes in primary earner status and anxiety/depression medication (eg, at year 1, odds ratio for women, 6.71 [95% CI, 1.96-23.01]; and for men, 0.69 [95% CI, 0.33-1.45]). However, except for anxiety/depression medication in women, similar changes were also observed in the general population.

CONCLUSIONS

OHCA survivors experience changes in employment, income, and primary earner status similar to the general population. However, women who survived OHCA more often received anxiety/depression medication in the years following OHCA.

Analysis of Longitudinal Binomial Data with Positive Association between the Number of Successes and the Number of Failures: An Application to Stock Instability Study

Numerous methods have been developed for longitudinal binomial data in the literature. These traditional methods are reasonable for longitudinal binomial data with a negative association between the number of successes and the number of failures over time; however, a positive association may occur between the number of successes and the number of failures over time in some behaviour, economic, disease aggregation and toxicological studies as the numbers of trials are often random. In this paper, we propose a joint Poisson mixed modelling approach to longitudinal binomial data with a positive association between longitudinal counts of successes and longitudinal counts of failures. This approach can accommodate both a random and zero number of trials. It can also accommodate overdispersion and zero inflation in the number of successes and the number of failures. An optimal estimation method for our model has been developed using the orthodox best linear unbiased predictors. Our approach not only provides robust inference against misspecified random effects distributions, but also consolidates the subject-specific and population-averaged inferences. The usefulness of our approach is illustrated with an analysis of quarterly bivariate count data of stock daily limit-ups and limit-downs.

Bringing pharmacomicrobiomics to the clinic through well-designed studies

Pharmacomicrobiomic studies investigate drug-microbiome interactions, such as the effect of microbial variation on drug response and disposition. Studying and understanding the interactions between the gut microbiome and drugs is becoming increasingly relevant to clinical practice due to its potential for avoiding adverse drug reactions or predicting variability in drug response. The highly variable nature of the human microbiome presents significant challenges to assessing microbes' influence. Studies aiming to explore drug-microbiome interactions should be well-designed to account for variation in the microbiome over time and collect data on confounders such as diet, disease, concomitant drugs, and other environmental factors. Here, we assemble a set of important considerations and recommendations for the methodological features required for performing a pharmacomicrobiomic study in humans with a focus on the gut microbiome. Consideration of these factors enable discovery, reproducibility, and more accurate characterization of the relationships between a given drug and the microbiome. Furthermore, appropriate interpretation and dissemination of results from well-designed studies will push the field closer to clinical relevance and implementation.

A Distribution-Free Model for Longitudinal Metagenomic Count Data

Longitudinal metagenomics has been widely studied in the recent decade to provide valuable insight for understanding microbial dynamics. The correlation within each subject can be observed across repeated measurements. However, previous methods that assume independent correlation may suffer from incorrect inferences. In addition, methods that do account for intra-sample correlation may not be applicable for count data. We proposed a distribution-free approach, namely CorrZIDF, which extends the current method to model correlated zero-inflated metagenomic count data, offering a powerful and accurate solution for detecting significance features. This method can handle different working correlation structures without specifying each margin distribution of the count data. Through simulation studies, we have shown the robustness of CorrZIDF when selecting a working correlation structure for repeated measures studies to enhance the efficiency of estimation. We also compared four methods using two real datasets, and the new proposed method identified more unique features that were reported previously on the relevant research.

Functional decline in the aphasic variant of Alzheimer's disease

INTRODUCTION

Primary progressive aphasia (PPA) is a clinical dementia syndrome associated with frontotemporal lobar degeneration (FTLD) or Alzheimer's disease (AD). Impairment in activities of daily living is essential for dementia diagnosis, yet less is known about the neuropathologic impact on functional decline in PPA, especially over time.

METHODS

Activities of Daily Living Questionnaire (ADLQ) ratings were compared by suspected underlying pathology between 17 PPA^Aβ+ and 11 PPA^Aβ- participants at 6-month intervals for 2 years using a linear mixed-effects model. A general linear model examined associations between functional decline and cortical thickness at baseline.

RESULTS

Groups did not differ in demographics or aphasia severity at baseline, yet overall and subdomain scores of the ADLQ were significantly worse for PPA^Aβ+ compared to PPA^Aβ- (P = .015) at each interval across 18 months.

DISCUSSION

Functional decline appears more pronounced and disrupts more aspects of life activities for individuals with non-semantic PPA with suspected AD versus non-AD neuropathology.

Power and sample size for GEE analysis of incomplete paired outcomes in 2 × 2 crossover trials

The 2 × 2 crossover trial uses subjects as their own control to reduce the intersubject variability in the treatment comparison, and typically requires fewer subjects than a parallel design. The generalized estimating equations (GEE) methodology has been commonly used to analyze incomplete discrete outcomes from crossover trials. We propose a unified approach to the power and sample size determination for the Wald Z-test and t-test from GEE analysis of paired binary, ordinal and count outcomes in crossover trials. The proposed method allows misspecification of the variance and correlation of the outcomes, missing outcomes, and adjustment for the period effect. We demonstrate that misspecification of the working variance and correlation functions leads to no or minimal efficiency loss in GEE analysis of paired outcomes. In general, GEE requires the assumption of missing completely at random. For bivariate binary outcomes, we show by simulation that the GEE estimate is asymptotically unbiased or only minimally biased, and the proposed sample size method is suitable under missing at random (MAR) if the working correlation is correctly specified. The performance of the proposed method is illustrated with several numerical examples. Adaption of the method to other paired outcomes is discussed.

The first-order Markov conditional linear expectation approach for analysis of longitudinal data

We consider longitudinal discrete data that may be unequally spaced in time and may exhibit overdispersion, so that the variance of the outcome variable is inflated relative to its assumed distribution. We implement an approach that extends generalized linear models for analysis of longitudinal data and is likelihood based, in contrast to generalized estimating equations (GEE) that are semiparametric. The method assumes independence between subjects; first-order antedependence within subjects; exponential family distributions for the first outcome on each subject and for the subsequent conditional distributions; and linearity of the expectations of the conditional distributions. We demonstrate application of the method in an analysis of seizure counts and in a study to evaluate the performance of transplant centers. Simulations for both studies demonstrate the benefits of the proposed likelihood based approach; however, they also demonstrate better than anticipated performance for GEE.

Distribution-free models for latent mixed population responses in a longitudinal setting with missing data

Many biomedical and psychosocial studies involve population mixtures, which consist of multiple latent subpopulations. Because group membership cannot be observed, standard methods do not apply when differential treatment effects need to be studied across subgroups. We consider a two-group mixture in which membership of latent subgroups is determined by structural zeroes of a zero-inflated count variable and propose a new approach to model treatment differences between latent subgroups in a longitudinal setting. It has also been incorporated with the inverse probability weighted method to address data missingness. As the approach builds on the distribution-free functional response models, it requires no parametric distribution model and thereby provides a robust inference. We illustrate the approach with both real and simulated data.

Use of a Mobile App to Facilitate Blood Glucose Monitoring in Adolescents With Type 1 Diabetes: Single-Subject Nonrandomized Clinical Trial

Background

Cloud-based glucose monitoring programs allow users with diabetes to wirelessly synchronize their glucometers to their mobile phones. They also provide visualization and remote access of their data through its mobile app. There have been very few studies evaluating their effectiveness in managing diabetes among adolescents with type 1 diabetes (T1D).

Objective

The purpose of this study was to assess the feasibility of using a mobile app to improve daily average blood glucose (BG) levels and increase BG monitoring frequency.

Methods

We used an ABA single-subject prospective study design. We recruited five participants aged 13 to 17 years with uncontrolled T1D, glycated hemoglobin A1c 9.0%-10.7%, self-monitoring behavior of ≤5 checks/day, and on multiple daily insulin injections. The study consisted of 4-week intervals of three phases: (1) phase A: usual glucose monitoring log (fax); (2) phase B: mobile app; and (3) phase A': second phase A. A certified diabetes educator and endocrinologist reviewed logs and provided recommendations weekly. Data were analyzed using a quasi-Poisson model to adjust for overdispersion among individual participants, and a generalized estimating equation model for overall intervention effect in aggregate.

Results

For mean daily BG (mg/dL) levels, participant 1 had decreased values on the mobile app (298 to 281, P=.03) and maintained in phase A'. Participant 4 had an increase in mean daily BG in phase A' (175 to 185, P=.01), whereas participant 5 had a decrease in mean daily BG in phase A' (314 to 211, P=.04). For daily monitoring (checks/day), participant 3 increased in phase B (4.6 to 8.3, P=.01) and maintained in phase A'. Participant 5 also had increased daily monitoring at each phase (2.1 to 2.4, P=.01; 2.4 to 3.4, P=.02). For the five participants combined, the overall mean BG and BG checks per day in phase A were mean 254.8 (SD 99.2) and mean 3.6 (SD 2.0), respectively, mean 223.1 (SD 95.7) and mean 4.5 (SD 3.0) in phase B, and mean 197.5 (SD 81.3) and mean 3.7 (SD 2.1) in phase A'. Compared to phase A, mean glucose levels declined during phase B and remained lower during phase A' (P=.002). There was no overall change in BG checks by phase (P=.25). However, mean BG levels negatively correlated with daily BG checks (r=–.47, P<.001). Although all participants had positive opinions about the app, its utilization was highly variable.

Conclusions

We demonstrated modest feasibility of adolescents with uncontrolled T1D utilizing a glucose monitoring mobile app. Further study is needed to better determine its effects on BG level and monitoring frequency. Psychosocial factors and motivational barriers likely influence adoption and continuous use of technology for diabetes management.

Hypothesis Testing and Statistical Analysis of Microbiome

After the initiation of Human Microbiome Project in 2008, various biostatistic and bioinformatic tools for data analysis and computational methods have been developed and applied to microbiome studies. In this review and perspective, we discuss the research and statistical hypotheses in gut microbiome studies, focusing on mechanistic concepts that underlie the complex relationships among host, microbiome, and environment. We review the current available statistic tools and highlight recent progress of newly developed statistical methods and models. Given the current challenges and limitations in biostatistic approaches and tools, we discuss the future direction in developing statistical methods and models for the microbiome studies.

"You Should Have Seen the Look on Your Face…": Self-awareness of Facial Expressions

The awareness of facial expressions allows one to better understand, predict, and regulate his/her states to adapt to different social situations. The present research investigated individuals' awareness of their own facial expressions and the influence of the duration and intensity of expressions in two self-reference modalities, a real-time condition and a video-review condition. The participants were instructed to respond as soon as they became aware of any facial movements. The results revealed that awareness rates were 57.79% in the real-time condition and 75.92% in the video-review condition. The awareness rate was influenced by the intensity and (or) the duration. The intensity thresholds for individuals to become aware of their own facial expressions were calculated using logistic regression models. The results of Generalized Estimating Equations (GEE) revealed that video-review awareness was a significant predictor of real-time awareness. These findings extend understandings of human facial expression self-awareness in two modalities.

On performance of parametric and distribution-free models for zero-inflated and over-dispersed count responses

Zero-inflated Poisson (ZIP) and negative binomial (ZINB) models are widely used to model zero-inflated count responses. These models extend the Poisson and negative binomial (NB) to address excessive zeros in the count response. By adding a degenerate distribution centered at 0 and interpreting it as describing a non-risk group in the population, the ZIP (ZINB) models a two-component population mixture. As in applications of Poisson and NB, the key difference between ZIP and ZINB is the allowance for overdispersion by the ZINB in its NB component in modeling the count response for the at-risk group. Overdispersion arising in practice too often does not follow the NB, and applications of ZINB to such data yield invalid inference. If sources of overdispersion are known, other parametric models may be used to directly model the overdispersion. Such models too are subject to assumed distributions. Further, this approach may not be applicable if information about the sources of overdispersion is unavailable. In this paper, we propose a distribution-free alternative and compare its performance with these popular parametric models as well as a moment-based approach proposed by Yu et al. [Statistics in Medicine 2013; 32: 2390-2405]. Like the generalized estimating equations, the proposed approach requires no elaborate distribution assumptions. Compared with the approach of Yu et al., it is more robust to overdispersed zero-inflated responses. We illustrate our approach with both simulated and real study data.

Full and home smoking ban adoption after a randomized controlled trial targeting secondhand smoke exposure reduction

INTRODUCTION

The current study examined home and full (i.e., home plus car) smoking ban adoption as secondary outcomes to a randomized controlled trial targeting reduced secondhand smoke exposure (SHSe) for children under treatment for cancer.

METHODS

Families with at least 1 adult smoker who reported SHSe for their children (n = 119) were randomized to control or intervention conditions and followed for 1 year with 5 assessments. Both groups were advised of the negative health outcomes associated with SHSe; the intervention group provided more in-depth counseling from baseline to 3 months. Parents reported on household and car smoking behavior, demographic, psychosocial, and medical/treatment-related information.

RESULTS

Regardless of group assignment, there was an increase in home (odds ration [OR] = 1.16, p = .074) and full (OR = 1.37, p = .001) smoking ban adoption across time. Families in the intervention group were more likely to adopt a full ban by 3 months, but this difference was nonsignificant by 12 months. Married parents (OR = 2.33, p = .006) and those with higher self-efficacy for controlling children's SHSe (OR = 1.11, p = .023) were more likely to have a home smoking ban; parents who reported smoking fewer cigarettes were more likely to adopt a home (OR = 1.62, p < .0001) or full (OR = 7.32, p = .038) ban.

CONCLUSIONS

Smoking bans are in-line with Healthy People 2020's tobacco objectives and may be more feasible for parents with medically compromised children for immediate SHSe reduction. Furthermore, interventions targeting full smoking bans may be a more effective for comprehensive elimination of SHSe.

Two Studies of Connectedness to Parents and Suicidal Thoughts and Behavior in Children and Adolescents

We tested hypotheses that greater connectedness to parent(s) is associated with lower risk for nonlethal suicidal thoughts and behavior (STB), termed direct protective effects, and that parent connectedness serves to moderate (lower) the risk for STB associated with psychopathology including major depressive episode (MDE), termed moderating protective effects. Independent samples of children and adolescents recruited for a multicenter study of familial alcoholism were studied. Generalized estimating equation models were used that adjusted for age, sex, and youth psychopathology variables. The sample for Study 1 was assessed at baseline and about 2- and 4-year follow-ups, with baseline characteristics of n = 921, M age = 14.3 ± 1.8 years, and 51.8% female. The sample for Study 2 was assessed at baseline and about 5-year follow-up, with baseline characteristics of n = 867, M age = 12.0 ± 3.2 years, and 51.0% female. In both studies, increased perceived connectedness to father but not mother was associated with lower risk for measures of STB, consistent with direct protective effects. In Study 1, measures of parent connectedness were associated with lower risk for STB but only for youth that did not experience MDE (or alcohol use disorder), inconsistent with moderating protective effects. Study 2 showed that connectedness to fathers was associated with lower risk for suicide plans or attempts (severe STB) but not frequent thoughts of death or dying (nonsevere STB). Improved connectedness to fathers may lower risk for STB in children and adolescents, consistent with direct protective effects. Hypotheses about moderating protective effects were not supported.

Ototoxicity in children with high-risk neuroblastoma: prevalence, risk factors, and concordance of grading scales--a report from the Children's Oncology Group

PURPOSE

Platinum-based therapy is the mainstay for management of high-risk neuroblastoma. Prevalence of platinum-related ototoxicity has ranged from 13% to 95% in previous reports; variability is attributable to small samples and disparate grading scales. There is no consensus regarding optimal ototoxicity grading. Furthermore, prevalence and predictors of hearing loss in a large uniformly treated high-risk neuroblastoma population are unknown. We address these gaps in our study.

PATIENTS AND METHODS

Audiologic testing was completed after administration of cisplatin alone (< 400 mg/m(2); exposure one) or after cisplatin (400 mg/m(2)) plus carboplatin (1,700 mg/m(2); exposure two). Hearing loss was graded using four scales (American Speech-Language-Hearing Association; Brock; Chang; and Common Terminology Criteria for Adverse Events, version 3 [CTCAEv3]).

RESULTS

Of 489 eligible patients, 333 had evaluable audiologic data. Median age at diagnosis was 3.3 years. Prevalence of severe hearing loss differed by scale. For those in the exposure-one group, prevalence ranged from 8% per Brock to 47% per CTCAEv3 (Brock v CTCAEv3 and Chang, P < .01; CTCAEv3 v Chang, P = .16); for those in the exposure-two group, prevalence ranged from 30% per Brock to 71% per CTCAEv3 (all pair-wise comparisons, P < .01). In patients requiring hearing aids, hearing loss was graded as severe in 49% (Brock), 91% (Chang), and 100% (CTCAEv3). Risk factors for severe hearing loss included exposure to cisplatin and carboplatin compared with cisplatin alone and hospitalization for infection.

CONCLUSION

Severe hearing loss is prevalent among children with high-risk neuroblastoma. Exposure to cisplatin combined with myeloablative carboplatin significantly increases risk. The Brock scale underestimates severe hearing loss and should be used with caution in this setting.

On assessing model fit for distribution-free longitudinal models under missing data

The generalized estimating equation (GEE), a distribution-free, or semi-parametric, approach for modeling longitudinal data, is used in a wide range of behavioral, psychotherapy, pharmaceutical drug safety, and healthcare-related research studies. Most popular methods for assessing model fit are based on the likelihood function for parametric models, rendering them inappropriate for distribution-free GEE. One rare exception is a score statistic initially proposed by Tsiatis for logistic regression (1980) and later extended by Barnhart and Willamson to GEE (1998). Because GEE only provides valid inference under the missing completely at random assumption and missing values arising in most longitudinal studies do not follow such a restricted mechanism, this GEE-based score test has very limited applications in practice. We propose extensions of this goodness-of-fit test to address missing data under the missing at random assumption, a more realistic model that applies to most studies in practice. We examine the performance of the proposed tests using simulated data and demonstrate the utilities of such tests with data from a real study on geriatric depression and associated medical comorbidities.

Distribution-free models for longitudinal count responses with overdispersion and structural zeros

Overdispersion and structural zeros are two major manifestations of departure from the Poisson assumption when modeling count responses using Poisson log-linear regression. As noted in a large body of literature, ignoring such departures could yield bias and lead to wrong conclusions. Different approaches have been developed to tackle these two major problems. In this paper, we review available methods for dealing with overdispersion and structural zeros within a longitudinal data setting and propose a distribution-free modeling approach to address the limitations of these methods by utilizing a new class of functional response models. We illustrate our approach with both simulated and real study data.