Latent Class Growth Modelling: A Tutorial

Lexical cohesion development in English as a foreign language learners' argumentative writing: A latent class growth model approach

This study aimed to estimate the latent frailty trajectories and identify corresponding predictors (sociodemographic, HIV-related, comorbidities, and behavioral) among a cohort of people with HIV (PWH). Longitudinal observational study using latent class growth modeling. Nine hundred and seventy-six PWH aged 40 years and older with frailty measured from at least two visits within the ACTG HAILO cohort were included. Frailty components included weakness, physical activity, weight loss, exhaustion, and slowness. Latent class growth models were estimated to capture change in frailty over time; multinomial logistic regression was used to estimate associations between predictors and frailty trajectory class. At baseline, participants were M = 51.5 years old ( SD = 7.5), 81% male ( n = 783), 48% White non-Hispanic ( n = 461), and 20% Hispanic ( n = 195). Latent class growth models identified three frailty trajectories: sustained robustness ( n = 811; 83%), worsening frailty ( n = 79; 8%), and frailty improvement ( n = 86; 9%). Older age, race, sex at birth, select comorbidities (cardiovascular disease, depression, type 2 diabetes), and behavioral characteristics (physical activity, smoking, and alcohol) were associated with fluctuations in frailty trajectories over time ( P < 0.05). Modifiable factors such as managing comorbidities and promoting physical activity present ideal opportunities for future interventions to prevent or slow the progression of frailty.

Popular longitudinal models allow for prediction of growth trajectories in alternative ways. In latent class growth models (LCGMs), person-level covariates predict membership in discrete latent classes that each holistically define an entire trajectory of change (e.g., a high-stable class vs. late-onset class vs. moderate-desisting class). In random coefficient growth models (RCGMs, also known as latent curve models), however, person-level covariates separately predict continuously distributed latent growth factors (e.g., an intercept vs. slope factor). This article first explains how complex and nonlinear interactions between predictors and time are recovered in different ways via LCGM versus RCGM specifications. Then a simulation comparison illustrates that, aside from some modest efficiency differences, such predictor relationships can be recovered approximately equally well by either model—regardless of which model generated the data. Our results also provide an empirical rationale for integrating findings about prediction of individual change across LCGMs and RCGMs in practice.

The purpose of this study was to investigate the trajectory of depressive symptoms in multicultural adolescents using longitudinal data, and to identify predictive factors related to depressive symptoms of multicultural adolescents using latent class analysis. We used six time-point data derived from the 2012 to 2017 Multicultural Adolescents Panel Study (MAPS). Latent growth curve modeling was used to assess the overall features of depressive symptom trajectories in multicultural adolescents, and latent class growth modeling was used to determine the number and shape of trajectories. We applied multinomial logistic regression analysis to each class to explore predictive factors. We found that the overall slope of depressive symptoms in multicultural adolescents increased. Latent class analysis demonstrated three classes: (1) high-increasing class (i.e., high intercept, significantly increasing slope), (2) moderate-increasing class (i.e., moderate intercept, significantly increasing slope), and (3) low-stable class (i.e., low intercept, no significant slope). In particular, we found that the difference in the initial intercept of depressive symptoms determined the subsequent trajectory. There is a need for early screening for depressive symptoms in multicultural adolescents and preparing individual mental health care plans.

Objectives: This study aimed to classify latent groups based on the developmental trajectories of prosocial behavior in early childhood. In addition, it aimed to verify whether peer play interactions (play interaction, play disruption, and play disconnection) and self-control during preschool years have significant influences on the classification of latent groups.Methods: This study included 1,201 children(617 males, 51.4%; 584 females, 48.6%) drawn from the 7th, and 9th-11th waves of the Panel Study on Korean Children (PSKC). Latent class growth modeling was used to analyze the data.Results: The patterns of changes in prosocial behavior in early childhood were classified into three groups: high-increasing, moderate-stable, and low-decreasing. Among the predictors, play interaction, play disruption, and self-control significantly affected the classification of the latent groups. Specifically, children who exhibited high levels of play interaction and low levels of play disruption during the preschool years were more likely to belong to the high-increasing group, indicating a more positive prosocial behavior. However, those with low self-control were more likely to belong to the low-decreasing group, showing a relatively negative developmental trajectory of prosocial behavior.Conclusion: This study confirmed heterogeneities in the developmental trajectories of prosocial behavior in early childhood, differentially predicted by preschool-age peer interaction qualities and self-regulation. The findings highlight the importance of taking a comprehensive approach in early childhood education to enhance children's prosocial behavior development utilizing both interpersonal and intrapersonal factors during preschool years.

BackgroundThe use and sharing of patient-generated health data (PGHD) by clinicians or researchers is expected to enhance the remote monitoring of specific behaviors that affect patient health. In addition, PGHD use could support patients’ decision-making on preventive care management, resulting in reduced medical expenses. However, sufficient evidence on the use and sharing of PGHD is lacking, and the impact of PGHD recording on patients’ health behavior changes remains unclear.ObjectiveThis study aimed to assess patients’ engagement with PGHD recording and to examine the impact of PGHD recording on their health behavior changes.MethodsThis supplementary analysis used the data of 47 postpartum women who had been assigned to the intervention group of our previous study for managing urinary incontinence. To assess the patients’ engagement with PGHD recording during the intervention period (8 weeks), the fluctuation in the number of patients who record their PGHD (ie, PGHD recorders) was evaluated by an approximate curve. In addition, to assess adherence to the pelvic floor muscle training (PFMT), the weekly mean number of pelvic floor muscle contractions performed per day among 17 PGHD recorders was examined by latent class growth modeling (LCGM).ResultsThe fluctuation in the number of PGHD recorders was evaluated using the sigmoid curve formula (R2=0.91). During the first week of the intervention, the percentage of PGHD recorders was around 64% (30/47) and then decreased rapidly from the second to the third week. After the fourth week, the percentage of PGHD recorders was 36% (17/47), which remained constant until the end of the intervention. When analyzing the data of these 17 PGHD recorders, PFMT adherence was categorized into 3 classes by LCGM: high (7/17, 41%), moderate (3/17, 18%), and low (7/17, 41%).ConclusionsThe number of PGHD recorders declined over time in a sigmoid curve. A small number of users recorded PGHD continuously; therefore, patients’ engagement with PGHD recording was low. In addition, more than half of the PGHD recorders (moderate- and low-level classes combined: 10/17, 59%) had poor PFMT adherence. These results suggest that PGHD recording does not always promote health behavior changes.

To explore the trajectory and influencing factors of kinetophobia in elderly patients with limb fracture during the rehabilitation period. In this retrospective study, we retrieved the follow-up records of 150 elderly patients with limb fractures from our hospital's electronic medical record system. We collected the demographic data and Tampa Scale for Kinesiophobia (TSK) scores of patients at postoperative day 1 (T0), 1 week (T1), 3 weeks (T2), 6 weeks (T3), and 12 weeks (T4) to track changes in kinesiophobia over time. We used Mplus 8.3 software to fit the development trajectory types of kinesiophobia based on TSK scores at time points T0 to T4 using a Latent Class Growth Model (LCGM). After selecting the best fitting model, logistic regression analysis was performed to identify the risk factors for kinesiophobia in different types. Receiver operating characteristic (ROC) curve and area under the curve (AUC) were used to compare the predictive value of relevant influencing factors for kinetophobia in elderly patients recovering from limb fracture. The TSK scores decreased steadily from T0 to T4 [(46.03±7.88) at T0, (41.14±8.89) at T1, (34.61±5.64) at T2, (29.95±6.79) at T3, and (26.71±5.03) at T4], [F (4, 745) = 193.1, P < 0.001]. We identified the trajectory of changes in kinesiophobia symptoms through LCGM, gradually establishing models with 1 to 5 categories. By integrating the results of relevant fit indices, we ultimately selected the best fitting model with 2 categories. Among them, 119 patients in Class 1 (79.3%) showed a slow and continuous decline in kinesiophobia symptoms from T0 to T4, while 31 patients in Class 2 (20.7%) exhibited rapid decline followed by rebound in kinesiophobia symptoms. Logistic regression showed that older the age (OR = 1.219), per capita monthly income < 3000 yuan (OR = 12.657), numeric rating scale (NRS), patients with higher NRS (OR = 2.401) and higher self-efficacy (OR = 1.212) were more likely to be in Class 1. The ROC curve results show that the combined above indicators have a higher predictive value for the changes in fear of movement in elderly patients with lower limb fractures during the rehabilitation period (AUC = 0.934), compared to age (AUC = 0.694), per capita monthly income (AUC = 0.654), NRS score (AUC = 0.812), and self-efficacy (AUC = 0.811) as individual indicators. As the recovery time progresses for elderly patients with limb fractures, the overall trend of kinesiophobia scores decreases. Kinesiophobia presents with two different trajectories, with age, average monthly income, NRS score, and self-efficacy being important factors influencing the trajectory categories of kinesiophobia changes.

Primary open angle glaucoma (POAG) is a chronic, progressive, irreversible, and potentially blinding optic neuropathy. The risk of blindness due to progressive visual field (VF) loss varies substantially from patient to patient. Early identification of those patients destined to rapid progressive visual loss is crucial to prevent further damage. In this article, a latent class growth model (LCGM) was developed to predict the binary outcome of VF progression using longitudinal mean deviation (MD) and pattern standard deviation (PSD). Specifically, the trajectories of MD and PSD were summarized by a functional principal component (FPC) analysis, and the estimated FPC scores were used to identify subgroups (latent classes) of individuals with distinct patterns of MD and PSD trajectories. Probability of VF progression for an individual was then estimated as weighted average across latent classes, weighted by posterior probability of class membership given baseline covariates and longitudinal MD/PSD series. The model was applied to the participants with newly diagnosed POAG from the Ocular Hypertension Treatment Study (OHTS), and the OHTS data was best fit by a model with 4 latent classes. Using the resultant optimal LCGM, the OHTS participants with and without VF progression could be accurately differentiated by incorporating longitudinal MD and PSD.

Background: There are 3 critical periods for developing obesity during childhood: the fetal period; a period of adiposity rebound between 4-6 years; and during adolescence. The approach of latent class growth ascertains the most interpretable set of classes. The resulting categorical latent variable represents the groups of homogenous individuals within the class to which they belong. Aim: We hypothesized that the early infantile period is another critical period at risk for developing obesity in adolescents using latent class growth models based on longitudinal height and weight data. Methods: The study cohort included 913 healthy adolescent volunteers (396 males and 517 females) aged 15-18 years in Japan. Complete data sets for height and weight at birth, 1.5, 3, 6, 9 and 12 years of age, as well as at present, were obtained. Body mass index (BMI) and age- and sex-specific BMI Z scores were used as measures of obesity. Adiposity rebound was defined as the point of minimal BMI during infancy. Three-group latent class growth models were used in the present study. Results: The BMI trajectories showed that individuals who consistently increased in BMI during infancy, i.e., they showed no adiposity rebound, were at the highest risk for adolescent obesity in both sexes. The Z score trajectories revealed that male subjects who were born with high Z scores and whose scores remained at high levels, as well as those who were born with low Z scores and whose scores had a sharp increase from birth to 1.5 years, were at risk for developing obesity. Female subjects who were born with relatively high Z scores and whose scores had a sharp increase from birth to 1.5 years were at risk. A representative figure is shown as Fig 1. Conclusion: These data suggest that the early infantile period is important for preventing further development of obesity. Further more detailed studies are required to identify individuals and the exact timing of the critical periods during early infancy to prevent later obesity.

Insomnia displays heterogeneous trajectories across adolescence, which may induce addictive behaviours, including internet gaming disorder and substance use. This study aimed to investigate the latent trajectory classes of insomnia symptoms over 2 years and to examine the associations between insomnia trajectories and these addictive behaviours. Participants were 910 adolescents from six middle schools in Shanghai, China (52.7% males; mean age=13.17 years). The three-wave survey measured insomnia symptoms, internet gaming disorder, substance use, depressive symptoms, and sociodemographic characteristics from 7th to 9th grade. Latent class growth modelling was performed to identify the latent trajectory classes of insomnia symptoms. Then multivariable logistic regressions were conducted within the best-fitting latent class growth model to examine the associations of insomnia trajectories with internet gaming disorder and substance use. Two latent trajectory classes of insomnia symptoms were recognised: the non-insomnia group (71.8%) and the insomnia group (28.2%). In the multivariable analysis controlling for baseline demographic variables and depressive symptoms, the insomnia group had a higher risk of developing internet gaming disorder (OR=2.203 [95% CI: 1.258-3.858]) and substance use (OR=2.215 [95% CI: 1.324-3.705]) compared with the non-insomnia group. These findings add to a growing body of research on heterogeneous trajectories of insomnia symptoms during adolescence, suggesting that intervention strategies are needed to target the characteristics or developmental patterns of different insomnia subgroups. The ultimate goal is to mitigate the impact of insomnia symptoms on adolescent addictive behaviours.

Objective This study employs the Latent Class Growth Model(LCGM) to identify symptom trajectory categories in patients with chronic respiratory failure(CRF)and examines the association between distinct trajectory categories and readmission risk. Methods A retrospective cohort study was conducted, encompassing CRF patients hospitalized at a tertiary comprehensive hospital between January 2022 and December 2024.Data on demographic characteristics, clinical indicators, symptom scores, and readmission rates were collected.LCGM was applied to model symptom trajectories and identify latent trajectory categories.Additionally, multivariate Cox proportional hazards models were utilized to examine the relationship between different trajectory categories and readmission risk. Results A total of 512 CRF patients were included.LCGM analysis identified three distinct symptom trajectory categories: stable, fluctuating, and deteriorating. Patients in the stable category exhibited minimal variations in symptom scores over time, while those in the fluctuating category demonstrated periodic fluctuations.Conversely, patients in the deteriorating category experienced a continuous escalation in symptom scores. Multivariate Cox regression analysis revealed that, compared to the stable trajectory, patients classified under the fluctuating and deteriorating categories exhibited significantly increased readmission risks, with hazard ratios of 1.56 and 2.34, respectively. Moreover, patients in the deteriorating category were found to have substantially longer hospital stays and significantly higher medical expenditures than those in the other two categories. Conclusion This study successfully identified three distinct symptom trajectory categories in CRF patients through the application of LCGM and established that fluctuating and deteriorating symptom patterns were significantly correlated with an increased risk of readmission. The findings highlight the necessity of prioritizing clinical attention toward patients exhibiting fluctuating and deteriorating symptom trajectories, implementing early interventions, and adopting personalized management strategies to mitigate readmission rates and enhance patient outcomes.

The purpose of this study was to explore whether subgroups of patients with different functional recovery trajectories after total hip arthroplasty can be discerned, as well as their predictors, using data from the Dutch Arthroplasty Register (Landelijke Registratie Orthopedische Implantaten [LROI]). We retrospectively reviewed prospectively collected Oxford Hip Scores (OHS) up to 1 year postoperatively for patients who had undergone a primary total hip arthroplasty. Latent class growth modeling was used to classify subgroups of patients according to the trajectory of functional recovery represented by the patients' OHS. We used multivariable multinomial logistic regression analysis to explore factors associated with class membership. A total of 6,030 patients were analyzed. Latent class growth modeling identified fast starters (fast initial improvement, high 12-month scores; 87.7%), slow starters (no initial change and subsequent improvement; 4.6%), and late dippers (initial improvement and subsequent deterioration; 7.7%). Factors associated with slow starters were female sex (odds ratio [OR], 1.63 [95% confidence interval (CI), 1.14 to 2.33]) and smoking (OR, 1.95 [95% CI, 1.26 to 3.03]); an anterior approach (OR, 0.47 [95% CI, 0.29 to 0.78]) had a protective effect against a less favorable response. Factors associated with late dippers were age of >75 years (OR, 1.62 [95% CI, 1.22 to 2.15]), smoking (OR, 1.68 [95% CI, 1.17 to 2.42]), American Society of Anesthesiologists (ASA) grade of III or IV (OR, 1.41 [95% CI, 1.05 to 1.91]), obesity (OR, 1.96 [95% CI, 1.43 to 2.69]), poorer EuroQol-5 Dimensions (EQ-5D) Self-Care (OR, 1.41 [95% CI, 1.09 to 1.82] for "some problems" and OR, 2.90 [95% CI, 1.39 to 6.03] for "unable"), poorer EQ-5D Anxiety/Depression (OR, 1.31 [95% CI, 1.00 to 1.71] for "moderately" and OR, 1.86 [95% CI, 1.06 to 3.24] for "extremely"), poorer EQ-5D visual analog scale (OR, 0.91 [95% CI, 0.86 to 0.97] per 10 points), direct lateral approach (OR, 2.18 [95% CI, 1.58 to 3.02]), and hybrid fixation with a cemented acetabular implant (OR, 1.79 [95% CI, 1.00 to 3.21]). We discerned fast starters, slow starters, and late dippers after total hip arthroplasty. Female sex, older age, obesity, higher ASA grades, and worse EQ-5D scores were associated with a less favorable response to total hip arthroplasty, as well as hybrid fixation (cemented acetabular implant) and direct lateral approach. Anterior approach had a protective effect against a less favorable response. However, all subgroups experienced functional improvement following total hip arthroplasty. Prognostic Level III. See Instructions for Authors for a complete description of levels of evidence.

This study examined the heterogeneity in BMI development by identifying distinct developmental trajectories. These trajectories were further investigated by relating them to markers of low-grade inflammation later in life. Data from approximately 400 healthy volunteers participating in the Spokane Heart Study were collected in 2-year intervals, and four waves of data were available for the current analyses. Body weight was measured by BMI and low-grade inflammation by high-sensitivity C-reactive protein (CRP) and fibrinogen. Up to date statistical techniques, i.e. latent class growth models, were used to analyse heterogeneity in body weight, and linear regressions were run to analyse possible associations between trajectories of body weight and CRP/fibrinogen levels. Six trajectories were identified (three stable, two increasing, and one decreasing) which differed significantly on CRP/fibrinogen levels, highlighting the importance of weight trajectories. The differences were only partly explained by variations in lifestyle habits.

Depression and vascular disease can both be risk factors for cognitive decline. This study assessed whether indicators of vascular disease are associated with depression over time, and whether depression is associated with cognition independent of vascular disease. Participants were 1032 community-dwelling, older adults from the Sydney Memory and Ageing Study. Depressive symptoms were measured on the 15-item Geriatric Depression Scale, and depression trajectories were classified using latent class growth modelling. Global cognition and individual cognitive domains were assessed. Indicators of vascular disease were history of stroke or transient ischaemic attack (TIA) and total white matter hyperintensities (TWMH). Generalised linear mixed modelling assessed whether vascular markers were associated with change in depressive symptoms over time. Linear mixed modelling examined relationships between depression, vascular disease indicators, and cognition. History of stroke/TIA or TWMH did not predict depression over time (p=0.11 and p=0.70). Baseline stroke/TIA history was associated with a decline in attention over time (b=-0.05, p=0.04). Significant differences between latent depression trajectories classes in cognitive decline were observed, with participants who became depressed over time experiencing decline in global cognition (b=-0.04, p<0.001), attention (b=-0.05, p<0.001) and executive function (b=-0.05, p<0.001), compared to the non-depressed group. Our findings suggest that depression in later life may not precede cognitive decline. Instead, depressive symptoms and cognitive decline, particularly in attention and executive function, may evolve in tandem over time. This reinforces the importance of early identification and treatment of depression in older adults to potentially mitigate cognitive deterioration.

ObjectiveGeneration Y has to earn a living in a new world of work in changing labor markets. To date, little is known about the working lives of todays’ young adults...