Missing data techniques for multilevel data: implications of model misspecification

Making Sense of Censored Covariates: Statistical Methods for Studies of Huntington's Disease

The landscape of survival analysis is constantly being revolutionized to answer biomedical challenges, most recently the statistical challenge of censored covariates rather than outcomes. There are many promising strategies to tackle censored covariates, including weighting, imputation, maximum likelihood, and Bayesian methods. Still, this is a relatively fresh area of research, different from the areas of censored outcomes (i.e., survival analysis) or missing covariates. In this review, we discuss the unique statistical challenges encountered when handling censored covariates and provide an in-depth review of existing methods designed to address those challenges. We emphasize each method's relative strengths and weaknesses, providing recommendations to help investigators pinpoint the best approach to handling censored covariates in their data.

A cotwin control study of associations between financial hardship and binge eating phenotypes during COVID-19

BACKGROUND

COVID-19 was associated with significant financial hardship and increased binge eating (BE). However, it is largely unknown whether financial stressors contributed to BE during the pandemic. We used a longitudinal, cotwin control design that controls for genetic/environmental confounds by comparing twins in the same family to examine whether financial hardship during COVID-19 was associated with BE.

METHODS

Female twins (N = 158; M_age  = 22.13) from the Michigan State University Twin Registry rated financial stressors (e.g., inability to afford necessities) daily for 49 consecutive days during COVID-19. We first examined whether financial hardship was associated with BE phenotypes across the full sample. We then examined whether cotwins who differed on financial hardship also differed in BE.

RESULTS

Participants who experienced greater mean financial hardship across the study had significantly greater dimensional BE symptoms, and participants who experienced greater financial hardship on a given day reported significantly more emotional eating that day. These results were replicated in cotwin control analyses. Twins who experienced more financial hardship than their cotwin across the study reported greater dimensional BE symptoms than their cotwin, and participants who experienced more financial hardship than their cotwin on a given day reported greater emotional eating that day. Results were identical when restricting analyses to monozygotic twins, suggesting associations were not due to genetic confounds.

CONCLUSIONS

Results suggest that BE-related symptoms may be elevated in women who experienced financial hardship during COVID-19 independent of potential genetic/environmental confounds. However, additional research in larger samples is needed.

PUBLIC SIGNIFICANCE

Little is known regarding how financial difficulties during the COVID-19 pandemic may have contributed to increased binge eating (BE). We found preliminary evidence that financial hardship during COVID-19 may be associated with greater rates of BE-related symptoms even when comparing twins from the same family. While additional research is needed, results suggest that people who experienced financial hardship during COVID-19 may be at increased risk for BE.

Adverse and Benevolent Childhood Experiences Predict Prenatal Sleep Quality

The objective of the study was to investigate whether adverse and benevolent childhood experiences were associated with trajectories of sleep quality throughout pregnancy. The study was conducted at obstetrics and gynecology clinics in the Rocky Mountain region of the USA. The participants of the study were pregnant individuals (N = 164). Sleep quality was measured with the Pittsburgh Sleep Quality Index at three gestational time points, and adverse childhood experiences (ACEs) and benevolent childhood experiences (BCEs) were assessed once. Multilevel models were conducted to examine the trajectory of sleep quality across gestation in relation to ACEs and BCEs. Sleep quality was similar in early to mid-pregnancy, with a worsening of sleep quality late in pregnancy, following a quadratic trajectory. Higher levels of ACEs predicted poorer prenatal sleep quality (b = 0.36, SE = 0.13, p = .004) throughout pregnancy, while higher levels of BCEs predicted better sleep quality (b = -0.60, SE = 0.17, p < .001) throughout pregnancy. Examination of ACEs subtypes revealed that childhood maltreatment predicted poor sleep quality (b = 0.66, SE = 0.18, p < .001), while childhood household dysfunction was not significantly associated (b = 0.33, SE = 0.21, p = .11). Associations remained after covarying for socioeconomic status and current stressful life events. Both adverse and benevolent childhood experiences predict sleep health during pregnancy. Prevention and intervention strategies targeting resilience and sleep quality during pregnancy should be implemented to promote prenatal health and well-being.

Longitudinal and prospective assessment of prenatal maternal sleep quality and associations with newborn hippocampal and amygdala volume

BACKGROUND

The rapid maturation of the fetal brain renders the fetus susceptible to prenatal environmental signals. Prenatal maternal sleep quality is known to have important health implications for newborns including risk for preterm birth, however, the effect on the fetal brain is poorly understood.

METHOD

Participants included 94 pregnant participants and their newborns (53% female). Pregnant participants (M_age = 30; SD_age= 5.29) reported on sleep quality three times throughout pregnancy. Newborn hippocampal and amygdala volumes were assessed using structural magnetic resonance imaging. Multilevel modeling was used to test the associations between trajectories of prenatal maternal sleep quality and newborn hippocampal and amygdala volume.

RESULTS

The overall trajectory of prenatal maternal sleep quality was associated with hippocampal volume (left: b = 0.00003, p = 0.013; right: b = 0.00003, p = .008). Follow up analyses assessing timing of exposure indicate that poor sleep quality early in pregnancy was associated with larger hippocampal volume bilaterally (e.g., late gestation left: b = 0.002, p = 0.24; right: b = 0.004, p = .11). Prenatal sleep quality was not associated with amygdala volume.

CONCLUSION

These findings highlight the implications of poor prenatal maternal sleep quality and its role in contributing to newborn hippocampal development.

Evaluation of approaches for accommodating interactions and non-linear terms in multiple imputation of incomplete three-level data

Three-level data structures arising from repeated measures on individuals clustered within larger units are common in health research studies. Missing data are prominent in such studies and are often handled via multiple imputation (MI). Although several MI approaches can be used to account for the three-level structure, including adaptations to single- and two-level approaches, when the substantive analysis model includes interactions or quadratic effects, these too need to be accommodated in the imputation model. In such analyses, substantive model compatible (SMC) MI has shown great promise in the context of single-level data. Although there have been recent developments in multilevel SMC MI, to date only one approach that explicitly handles incomplete three-level data is available. Alternatively, researchers can use pragmatic adaptations to single- and two-level MI approaches, or two-level SMC-MI approaches. We describe the available approaches and evaluate them via simulations in the context of three three-level random effects analysis models involving an interaction between the incomplete time-varying exposure and time, an interaction between the time-varying exposure and an incomplete time-fixed confounder, or a quadratic effect of the exposure. Results showed that all approaches considered performed well in terms of bias and precision when the target analysis involved an interaction with time, but the three-level SMC MI approach performed best when the target analysis involved an interaction between the time-varying exposure and an incomplete time-fixed confounder, or a quadratic effect of the exposure. We illustrate the methods using data from the Childhood to Adolescence Transition Study.

Infant temperamental fear, pupil dilation, and gaze aversion from smiling strangers

In childhood, higher levels of temperamental fear-an early-emerging proclivity to distress in the face of novelty-are associated with lower social responsivity and greater social anxiety. While the early emergence of temperamental fear in infancy is poorly understood, it is theorized to be driven by individual differences in reactivity and self-regulation to novel stimuli. The current study used eye tracking to capture infants' (N = 124) reactions to a video of a smiling stranger-a common social encounter-including infant gaze aversions from the stranger's face (indexing arousal regulation) and pupil dilation (indexing physiological reactivity), longitudinally at 2, 4, 6, and 8 months of age. Multilevel mixed-effects models indicated that more fearful infants took more time to look away from a smiling stranger's face than less fearful infants, suggesting that high-fear infants may have slower arousal regulation. At 2 and 4 months, more fearful infants also exhibited greater and faster pupil dilation before gaze aversions, consistent with greater physiological reactivity. Together, these findings suggest that individual differences in infants' gaze aversions and pupil dilation can index the development of fearful temperament in early infancy, facilitating the identification of, and interventions for, risk factors to social disruptions.

A daily diary study of emotion regulation as a moderator of negative affect-binge eating associations

BACKGROUND

While negative affect (NA) typically increases risk for binge eating, the ultimate impact of NA may depend on a person's ability to regulate their emotions. In this daily, longitudinal study, we examined whether emotion regulation (ER) modified the strength of NA-dysregulated eating associations.

METHODS

Women (N = 311) from the Michigan State University Twin Registry first reported dimensional binge eating symptoms and broad ER difficulties (e.g., limited emotional awareness, difficulty controlling emotional impulses). Participants then rated use of adaptive (cognitive reappraisal, social sharing, situation modification, and acceptance) and maladaptive (rumination, expressive suppression, and self-criticism) ER strategies, emotional eating (EE), objective binge eating (OBE), and NA once daily for 49 consecutive days.

RESULTS

There were several main effects of ER on binge-eating pathology in both between-person (i.e., comparing women who differed on average) and within-person (i.e., examining fluctuations in variables day-to-day) analyses. Between-person, greater broad ER difficulties, greater maladaptive strategy use, and lower adaptive strategy use were all associated with greater binge-eating pathology. Within-person, greater maladaptive strategy use was associated with greater odds of OBE on that day and on the following day. However, neither broad ER difficulties nor use of specific strategies moderated associations between NA and dysregulated eating in between- or within-person analyses.

CONCLUSIONS

While ER is independently associated with risk for dysregulated eating, it may not fully mitigate the impact of NA. Additional strategies (e.g., decreasing environmental stressors and increasing social support) may be needed to minimize NA and its impact on dysregulated eating.

PUBLIC SIGNIFICANCE

Negative affect (NA; e.g., sadness, guilt) increases dysregulated eating risk. Because NA is sometimes unavoidable, we examined whether emotion regulation (ER; i.e., how a person responds to their emotions) might impact whether NA leads to dysregulated eating. Although more effective ER was associated with less dysregulated eating overall, ER did not impact the association between NA and dysregulated eating. Other approaches may therefore be needed to mitigate NA-dysregulated eating associations.

Disparities in the prevalence, frequency and trajectories of substance use and disordered eating across first-year university in sexual minority undergraduates

Objective. To compare the prevalence and trajectories of risky health behaviors in sexual minority (SM; lesbian/gay/bisexual/questioning/asexual) versus heterosexual undergraduates across their first year of post-secondary. Participants. First-year undergraduates (N = 704, Mage=17.97 years, 24% SM) from a midsized Canadian university. Methods. Students completed monthly (September-April) online questionnaires assessing substance use (binge drinking, tobacco, cannabis, illicit drug use) and disordered eating (binge eating, fasting, purging). Results. At the outset of the academic year, SM students endorsed more frequent cannabis use, illicit drug use, fasting, and binge eating, but less frequent tobacco use, versus heterosexual students. Over the year, SM students' binge drinking frequency declined less than that of heterosexual students, but their illicit drug use decreased while that of their heterosexual peers increased, and all students reported declining disordered eating frequency. Conclusions. Campus wellness initiatives for SM students should offer prevention and harm-reduction strategies prior to or shortly after their arrival on campus.

Multiple imputation approaches for handling incomplete three-level data with time-varying cluster-memberships

Three-level data arising from repeated measures on individuals clustered within higher-level units are common in medical research. A complexity arises when individuals change clusters over time, resulting in a cross-classified data structure. Missing values in these studies are commonly handled via multiple imputation (MI). If the three-level, cross-classified structure is modeled in the analysis, it also needs to be accommodated in the imputation model to ensure valid results. While incomplete three-level data can be handled using various approaches within MI, the performance of these in the cross-classified data setting remains unclear. We conducted simulations under a range of scenarios to compare these approaches in the context of an acute-effects cross-classified random effects substantive model, which models the time-varying cluster membership via simple additive random effects. The simulation study was based on a case study in a longitudinal cohort of students clustered within schools. We evaluated methods that ignore the time-varying cluster memberships by taking the first or most common cluster for each individual; pragmatic extensions of single- and two-level MI approaches within the joint modeling (JM) and the fully conditional specification (FCS) frameworks, using dummy indicators (DI) and/or imputing repeated measures in wide format to account for the cross-classified structure; and a three-level FCS MI approach developed specifically for cross-classified data. Results indicated that the FCS implementations performed well in terms of bias and precision while JM approaches performed poorly. Under both frameworks approaches using the DI extension should be used with caution in the presence of sparse data.

The Magnetic Couples Study: protocol for a mixed methods prospective cohort study of HIV-serodifferent heterosexual couples' perspectives and use of pre-exposure prophylaxis (PrEP)

INTRODUCTION

HIV transmission within serodifferent heterosexual couples plays a key role in sustaining the global HIV pandemic. In the USA, transmission within established mixed-status couples accounts for up to half of all new HIV infections among heterosexuals. Oral HIV pre-exposure prophylaxis (PrEP) is a highly effective prevention method, although underutilised among serodifferent couples. Moreover, there is a dearth of research on US HIV-serodifferent couples' perspectives and use of PrEP, alone or in combination with other prevention methods. In this paper, we describe the study protocol for the Magnetic Couples Study, designed to fill critical knowledge gaps regarding HIV-serodifferent heterosexual couples' perspectives, experiences and utilisation of PrEP.

METHODS AND ANALYSIS

The Magnetic Couples Study is a mixed methods prospective cohort study designed to describe temporal patterns and identify determinants at multiple levels (individual, couple, HCF) of PrEP outcomes along the care continuum (PrEP awareness, linkage, uptake, retention and medication adherence) among HIV-serodifferent heterosexual couples residing in New York City. The study will also examine clinical management of PrEP, side effects and changes in sexual-related and substance use-related behaviour. A prospective cohort of 230 mixed-status couples already on oral PrEP was recruited, with quarterly assessments over 18 months; in addition, a cross-sectional sample of 150 mixed-status couples not currently on PrEP was recruited. In-depth semistructured qualitative interviews were conducted with a subsample of 25 couples. Actor-partner interdependence modelling using multilevel analysis will be employed for the analysis of longitudinal dyadic data. Framework analysis will be used to analyse qualitative data. A parallel convergent design will be used for mixed methods integration.

ETHICS AND DISSEMINATION

The study was approved by the University of Rochester Institutional Review Board (RSRB00052766). Study findings will be disseminated to community members and providers and to researchers and policy makers.

Longitudinal Associations of Conscientiousness and Neuroticism With Perceived Mastery and Constraints for Aging Adults

It has not been well understood how conscientiousness and neuroticism are associated with two related but distinct dimensions of perceived control (i.e., perceived mastery and constraints) among aging adults. The present study examined these associations and their change over time, while addressing whether they differ by age or gender. For respondents aged 50+ at baseline (N = 2,768) in the Midlife in the United States (MIDUS) study, multilevel modeling analyses were conducted to assess how conscientiousness and neuroticism predicted perceived mastery and constraints over 2 decades. As expected, higher conscientiousness and lower neuroticism (for both between- and within-person variability) predicted higher perceived mastery and lower perceived constraints overall. Nuanced findings emerged related to age, gender and change over time for different associations of conscientiousness and neuroticism with the outcomes. These findings can inform future research suggesting directions of further investigations for these complex associations.

Associations of emotional burden and coping strategies with sick leave among healthcare professionals: A longitudinal observational study

OBJECTIVES

To investigate 1) whether care-related regrets (regret intensity, number of recent regrets) are associated with sick leave, independently of personality traits, perceived safety climate, and physical activity; and 2) whether these associations were mediated or moderated by coping strategies.

METHODS

Using a longitudinal international observational study (ICARUS), data were collected by the means of a weekly web survey. Descriptive and generalized estimation equations were performed.

RESULTS

A total of 276 newly practicing healthcare professionals (nurses, physicians, others) from 11 countries were included in this study. The average proportion of weeks with at least one day of sick leave was 3.2%. Nurses' sick leave increased with number of care-related regrets (Relative Risk [RR]=1.52; 95% Confidence Interval [CI]=[1.18; 1.95], p=.001), while physicians' sick leave increased with intensity of care-related regret (RR=1.21; 95%CI=[1.00; 1.21], p=.049). Coping was associated with lower risk of sick leave for nurses (RR _{problem-focused strategies} = 0.53; 95%CI=[0.37; 0.74], p=.001, and RR_{physical activity}=0.68; 95%CI:[0.54; 0.85], p<.001), but not for physicians. Nevertheless, the association of regret with sick leave remained significant even when adjusting for coping. Finally, this study did not find evidence of moderation by the coping strategies.

CONCLUSION

Regrets are associated with increased risks of sick leave, even in young healthcare professionals. Use of coping strategies partially mediated these associations in nurses. The results of this study should be used to inform interventions to reduce emotional burdens and enhance protective coping strategies.

Evaluation of approaches for multiple imputation of three-level data

Background

Three-level data arising from repeated measures on individuals who are clustered within larger units are common in health research studies. Missing data are prominent in such longitudinal studies and multiple imputation (MI) is a popular approach for handling missing data. Extensions of joint modelling and fully conditional specification MI approaches based on multilevel models have been developed for imputing three-level data. Alternatively, it is possible to extend single- and two-level MI methods to impute three-level data using dummy indicators and/or by analysing repeated measures in wide format. However, most implementations, evaluations and applications of these approaches focus on the context of incomplete two-level data. It is currently unclear which approach is preferable for imputing three-level data.

Methods

In this study, we investigated the performance of various MI methods for imputing three-level incomplete data when the target analysis model is a three-level random effects model with a random intercept for each level. The MI methods were evaluated via simulations and illustrated using empirical data, based on a case study from the Childhood to Adolescence Transition Study, a longitudinal cohort collecting repeated measures on students who were clustered within schools. In our simulations we considered a number of different scenarios covering a range of different missing data mechanisms, missing data proportions and strengths of level-2 and level-3 intra-cluster correlations.

Results

We found that all of the approaches considered produced valid inferences about both the regression coefficient corresponding to the exposure of interest and the variance components under the various scenarios within the simulation study. In the case study, all approaches led to similar results.

Conclusion

Researchers may use extensions to the single- and two-level approaches, or the three-level approaches, to adequately handle incomplete three-level data. The two-level MI approaches with dummy indicator extension or the MI approaches based on three-level models will be required in certain circumstances such as when there are longitudinal data measured at irregular time intervals. However, the single- and two-level approaches with the DI extension should be used with caution as the DI approach has been shown to produce biased parameter estimates in certain scenarios.

Staff perception of Lean, care-giving, thriving and exhaustion: a longitudinal study in primary care

Background

Lean is commonly adopted in healthcare to increase quality of care and efficiency. Few studies of Lean involve staff-related outcomes, and few have a longitudinal design. Thus, the aim was to examine the extent to which changes over time in Lean maturity are associated with changes over time in care-giving, thriving and exhaustion, as perceived by staff, with a particular emphasis on the extent to which job demands and job resources, as perceived by staff, have a moderated mediation effect.

Method

A longitudinal study with a correlational design was used. In total, 260 staff at 46 primary care units responded to a web survey in 2015 and 2016. All variables in the study were measured using staff ratings. Ratings of Lean maturity reflect participants’ judgements regarding the entire unit; ratings of care-giving, thriving, exhaustion and job demands and resources reflect participants’ judgements regarding their own situation.

Results

First, over time, increased Lean maturity was associated with increased staff satisfaction with their care-giving and increased thriving, mediated by increased job resources. Second, over time, increased Lean maturity was associated with decreased staff exhaustion, mediated by decreased job demands. No evidence was found showing that job demands and job resources had a moderated mediation effect.

Conclusion

The results indicate that primary care staff may benefit from working in organizations characterized by high levels of Lean maturity and that caregiving may also be improved as perceived by staff.

Electronic supplementary material

The online version of this article (10.1186/s12913-019-4502-6) contains supplementary material, which is available to authorized users.

Energy Drinks and Binge Drinking Predict College Students' Sleep Quantity, Quality, and Tiredness

This study examines whether energy drink use and binge drinking predict sleep quantity, sleep quality, and next-day tiredness among college students. Web-based daily data on substance use and sleep were collected across four semesters in 2009 and 2010 from 667 individuals for up to 56 days each, yielding information on 25,616 person-days. Controlling for average levels of energy drink use and binge drinking (i.e., 4+ drinks for women, 5+ drinks for men), on days when students consumed energy drinks, they reported lower sleep quantity and quality that night, and greater next-day tiredness, compared to days they did not use energy drinks. Similarly, on days when students binge drank, they reported lower sleep quantity and quality that night, and greater next-day tiredness, compared to days they did not binge drink. There was no significant interaction effect between binge drinking and energy drink use on the outcomes.

The Effects of Including Observed Means or Latent Means as Covariates in Multilevel Models for Cluster Randomized Trials

We investigated methods of including covariates in two-level models for cluster randomized trials to increase power to detect the treatment effect. We compared multilevel models that included either an observed cluster mean or a latent cluster mean as a covariate, as well as the effect of including Level 1 deviation scores in the model. A Monte Carlo simulation study was performed manipulating effect sizes, cluster sizes, number of clusters, intraclass correlation of the outcome, patterns of missing data, and the squared correlations between Level 1 and Level 2 covariates and the outcome. We found no substantial difference between models with observed means or latent means with respect to convergence, Type I error rates, coverage, and bias. However, coverage could fall outside of acceptable limits if a latent mean is included as a covariate when cluster sizes are small. In terms of statistical power, models with observed means performed similarly to models with latent means, but better when cluster sizes were small. A demonstration is provided using data from a study of the Tools for Getting Along intervention.

Web-Based Acceptance and Commitment Therapy for Mental Health Problems in College Students: A Randomized Controlled Trial

There are significant challenges in addressing the mental health needs of college students. The current study tested an acceptance and commitment therapy (ACT), web-based self-help program to treat a broad range of psychological problems students struggle with. A sample of 79 college students was randomized to web-based ACT or a waitlist condition, with assessments at baseline and posttreatment. Results indicated adequate acceptability and program engagement for the ACT website. Relative to waitlist, participants receiving ACT improved on overall distress, general anxiety, social anxiety, depression, academic concerns, and positive mental health. There were no between-group effects on eating concerns, alcohol use, or hostility, or on some key ACT process of change measures. ACT participants improved more on mindful acceptance and obstruction to valued living, both of which mediated treatment outcomes. Results are discussed in the context of lessons learned with the website prototype, and areas for further research are presented.

Multiple imputation methods for handling missing data in cost-effectiveness analyses that use data from hierarchical studies: an application to cluster randomized trials

PURPOSE

Multiple imputation (MI) has been proposed for handling missing data in cost-effectiveness analyses (CEAs). In CEAs that use cluster randomized trials (CRTs), the imputation model, like the analysis model, should recognize the hierarchical structure of the data. This paper contrasts a multilevel MI approach that recognizes clustering, with single-level MI and complete case analysis (CCA) in CEAs that use CRTs.

METHODS

We consider a multilevel MI approach compatible with multilevel analytical models for CEAs that use CRTs. We took fully observed data from a CEA that evaluated an intervention to improve diagnosis of active labor in primiparous women using a CRT (2078 patients, 14 clusters). We generated scenarios with missing costs and outcomes that differed, for example, according to the proportion with missing data (10%-50%), the covariates that predicted missing data (individual, cluster-level), and the missingness mechanism: missing completely at random (MCAR), missing at random (MAR), or missing not at random (MNAR). We estimated incremental net benefits (INBs) for each approach and compared them with the estimates from the fully observed data, the "true" INBs.

RESULTS

When costs and outcomes were assumed to be MCAR, the INBs for each approach were similar to the true estimates. When data were MAR, the point estimates from the CCA differed from the true estimates. Multilevel MI provided point estimates and standard errors closer to the true values than did single-level MI across all settings, including those in which a high proportion of observations had cost and outcome data MAR and when data were MNAR.

CONCLUSIONS

Multilevel MI accommodates the multilevel structure of the data in CEAs that use cluster trials and provides accurate cost-effectiveness estimates across the range of circumstances considered.