The Hitchhiker’s Guide to Longitudinal Models: A Primer on Model Selection for Repeated-Measures Methods

Longitudinal data is becoming increasingly available in developmental neuroimaging. To maximize the promise of this wealth of information on how biology, behavior, and cognition change over time, there is a need to incorporate broad and rigorous training in longitudinal methods into the repertoire of developmental neuroscientists. Fortunately, these models have an incredibly rich tradition in the broader developmental sciences that we can draw from. Here, we provide a primer on longitudinal models, written in a beginner-friendly (and slightly irreverent) manner, with a particular focus on selecting among different modeling frameworks (e.g., multilevel versus latent curve models) to build the theoretical model of development a researcher wishes to test. Our aims are three-fold: 1) lay out a heuristic framework for longitudinal model selection, 2) build a repository of references that ground each model in its tradition of methodological development and practical implementation with a focus on connecting researchers to resources outside traditional neuroimaging journals, and 3) provide practical resources in the form of a codebook companion demonstrating how to fit these models. These resources together aim to enhance training for the next generation of developmental neuroscientists by providing a solid foundation for future forays into advanced modeling applications.

Longitudinal associations between borderline personality disorder and five-factor model traits over 24 years

Are five-factor traits and borderline personality symptoms the same features with different names? The existing literature offers reasons to think they are the same and reasons to think they are different. We examined longitudinal associations between these variables in a sample of patients assessed 12 times over 24 years using latent curve models with structured residuals. Mean trajectories for all variables were in the direction of symptom reduction/personality maturation and could be parsed into an initial, rapid improvement phase and a subsequent, gradual improvement phase. We found robust between-person associations among intercepts and long-term slopes of traits and symptoms. Specifically, higher levels of neuroticism as well as lower levels of extraversion, agreeableness, and conscientiousness were associated with higher levels of borderline personality symptoms, and changes in these traits were correlated with reduction in symptoms over time. Associations among time-structured residuals allowed for examinations of within-person deflections from these general trends at briefer (two year) intervals. All variables exhibited robust within-person carry-over effects. Other within-person effects were more specific to certain traits. These results suggest that, despite their distinct theoretical and methodological bases, normal trait and psychiatric diagnostic approaches largely converged on a similar conception of borderline personality.

Longitudinal associations between borderline personality disorder and five factor model traits over 24 years

Are five factor traits and borderline personality symptoms the same features with different names? The existing literature offers reasons to think they are the same and reasons to think they are different. We examined longitudinal associations between these variables in a sample of patients assessed 12 times over 24 years using latent curve models with structured residuals (LCM-SR). Mean trajectories for all variables were in the direction of symptom reduction/personality maturation, and could be parsed into an initial, rapid improvement phase and a subsequent, gradual improvement phase. We found robust between-person associations among intercepts and long-term slopes of traits and symptoms. Specifically, higher levels of neuroticism as well as lower levels of extraversion, agreeableness, and conscientiousness were associated with lower levels of borderline personality symptoms, and changes in these traits were correlated with reduction in symptoms over time. Associations among time-structured residuals allowed for examinations of within-person deflections from these general trends at briefer (two year) intervals. All variables exhibited robust within-person carry-over effects. Other within-person effects were more specific to certain traits. These results suggest that, despite their distinct theoretical and methodological bases, normal trait and psychiatric diagnostic approaches largely converged on a similar conception of borderline personality.

THE INFLUENCE OF LONELINESS AND RURAL RESIDENCE ON DEPRESSION IN LATER LIFE

Loneliness is associated with depression among older adults. Limited research has examined the role of rurality in relationship to loneliness and depression; the extant research has mixed findings. The socioemotional selectivity theory states that as people age the quality of relationships become more important than the quantity (English & Carstensen, 2016). Individuals in rural areas may have a low quantity of relationships but deeper social ties within the community; thus, they may be less likely to become depressed over time. The association between loneliness and depression may be amplified for people in non-rural areas because they are surrounded by other people but lack close relationships that are most important during the aging process. This study examines the effect of living in rural areas on loneliness on predicting baseline depression and loneliness, as well as changes in these outcomes over time. Data are from the 2006-2014 waves of Health Retirement Study. Regression models examine the relationship between depression loneliness and rural residence controlling for health conditions and demographic characteristics. Latent curve models examine the disparity in trajectories of loneliness and depressive symptoms by urban and rural residence. Older adults who feel lonely (p<.001) and in urban areas (p<.0.05) are more likely to be depressed. Furthermore, the effect of loneliness on depression is weakened by rural residence (p<.05). It is salient to understand the protective effect of rural residency on depression among older adults in the U.S. We discuss implications for policy.

Influential factors of postoperative pain trajectories in patients receiving intravenous patient-controlled analgesia: a single-centre cohort study in Taiwan

ObjectivesWe aimed to investigate the factors associated with variations in postoperative pain trajectories over time in patients using intravenous patient-controlled analgesia (IV-PCA) for postoperative pain.DesignRetrospective cohort study.SettingA single medical centre in Taiwan.ParticipantsPatients receiving IV-PCA after surgery.Primary and secondary outcome measuresPrimary outcome was the postoperative pain scores.ResultsA total of 3376 patients and 20 838 pain score observations were analysed using latent curve models. Female and longer anaesthesia time increased the baseline level of pain (p=0.004 and 0.003, respectively), but abdominal surgery and body weight decreased it (both p<0.001). Regarding the trend of pain resolution, lower abdominal surgery steepened the slope (p<0.001); older age, American Society of Anesthesiologists (ASA) class ≥3 and longer anaesthesia time tended to flatten the slope (p<0.001, =0.019 and <0.001, respectively). PCA settings did not affect the variations in postoperative pain trajectories.ConclusionsPatient demographics, ASA class, anaesthesia time and surgical sites worked together to affect postoperative pain trajectories in patients receiving IV-PCA. Latent curve models provided valuable information about the dynamic and complex relationships between the pain trajectories and their influential factors.

Mixture Modeling for Lifespan Developmental Research

As part of the Generalized Structural Equation Modeling framework, mixture models are person-centered analyses seeking to identify distinct subpopulations, or profiles, of participants differing quantitatively and qualitatively from one another on a configuration of indicators and/or relations among these indicators. Mixture models are typological (resulting in a classification system), probabilistic (each participant having a probability of membership into all profiles based on prototypical similarity), and exploratory (the optimal model is typically selected based on a comparison of alternative specifications) in nature, and can take different forms. Latent profile analyses seek to identify subpopulations of participants differing from one another on a configuration of indicators and can be extended to factor mixture analyses allowing for the incorporation of latent factors to the model. In contrast, mixture regression analyses seek to identify subpopulations of participants’ differing from one another in terms of relations among profile indicators. These analyses can be extended to the multiple-group and/or longitudinal analyses, allowing researchers to conduct tests of profile similarity across different samples of participants or time points, and latent transition analyses can be used to assess probabilities of profiles transition over time among a sample of participants (i.e., within person stability and change in profile membership). Finally, growth mixture analyses are built from latent curve models and seek to identify subpopulations of participants following quantitatively and qualitatively distinct trajectories over time. All of these models can accommodate covariates, used either as predictors, correlates, or outcomes, and can even be extended to tests of mediation and moderation.

Prescriber Variation in Relation to Prescribing Trends within the Preferred Drugs Initiative in Ireland (2012–2015): An Interrupted Time-Series Study Using Latent Curve Models

Objectives. To examine the impact of the Preferred Drugs Initiative (PDI), an Irish health policy aimed at reducing prescribing variation. Design. Interrupted time series spanning 2012 to 2015. Setting. Health Service Executive pharmacy claims data for General Medical Services (GMS) patients, approximately 40% of the Irish population. Participants. Prescribers issuing preferred drug group items to GMS adults before and after PDI guidelines. Primary Outcome. The percentage coverage of PDI medications within each drug class per calendar quarter per prescriber. Methods. Latent curve models with structured residuals (LCM-SRs) were used to model coverage of the preferred drugs over time. The number of GMS adults receiving medication and the percentage who were 65 years and older at the start of the study were included as covariates. Results. In the quarter following PDI guidelines, coverage of the preferred drugs increased most in absolute terms for proton pump inhibitors (PPIs) (1.50% [SE 0.15], P < 0.001) and selective and norepinephrine reuptake inhibitors (SNRIs) (1.17% [SE 0.26], P < 0.001). Variation between prescribers remained relatively unchanged and increased for urology medications. Prescribers who increased coverage of the preferred PPI also increased coverage of the preferred statin immediately following guidelines (correlation 0.47 [SE 0.13], P < 0.001). Where guidelines were disseminated simultaneously, coverage of one preferred drug did not significantly predict coverage of the other preferred drug in the next calendar quarter. Prescribing of preferred drugs was not moderated by prescriber-level factors. Conclusions. Modest changes in prescribing of the preferred drugs have been observed over the course of the PDI. However, the guidelines have had little impact in reducing variation between prescribers. Further strategies may be necessary to reduce variation in clinical practice and enhance patient care.