Joint Modeling Of Longitudinal Data Research Articles

Joint models for longitudinal data

In a longitudinal study, data on different types of variables are often collected repeatedly over time. Some variables may be continuous, and some variables may be binary or times to an event of interest. Even for a single variable, data may be collected at different phases of the study with different characteristics. These different types of variables are typically associated or correlated, since they are measurements on the same individuals in the study. Analysis of data on each of these variables separately, ignoring other variables, may be inefficient and may also lead to biased results. Standard multivariate models with several correlated responses may not be easy to specify for different types of variables or when the models are nonlinear. Jointly modelling these variables simultaneously not only may be more efficient but may also reduce biases in parameter estimation. Statistical inference can then be based on the joint likelihood for all observed data. In this article, we briefly review several different types of joint models for longitudinal data. We focus on mixed effects models and likelihood methods for inference. We illustrate these joint models with datasets from HIV/AIDS studies. Journal of Statistical Research 2024, Vol. 58, No. 1, pp. 75-96.

Systolic Blood Pressure Levels and Probable Dementia : Secondary Analysis of a Randomized Clinical Trial

AbstractBackgroundDespite a significant 19% reduction in risk for mild cognitive impairment, the primary Systolic Blood Pressure Intervention Trial (SPRINT) analysis lacked sufficient evidence to assert statistical significance for the 17% reduction in probable dementia alone in intensive blood pressure (BP) control, defined as a target systolic BP < 120 mmHg, compared to the standard target of < 140 mmHg (p = 0.10). However, early termination of SPRINT may have caused these analyses to be under‐powered. We applied the statistical framework of joint models for longitudinal data and time‐to‐event outcomes to analyze the effect of systolic BP levels on probable dementia using the SPRINT data.MethodsSPRINT was a multicenter randomized clinical trial that enrolled patients ≥50 years of age with hypertension and increased cardiovascular risk but without diabetes or history of stroke. In this secondary analysis, the primary outcome was occurrence of adjudicated probable dementia after randomization. Secondary outcomes included adjudicated mild cognitive impairment and a composite outcome of mild cognitive impairment or probable dementia. To analyze these outcomes, we applied joint longitudinal and survival models. In the longitudinal component of the joint model, systolic BP levels were modeled as a quadratic function of time since randomization, accounting for correlation in repeated measurements and for assignment to intensive BP control versus standard care. In the survival component, multi‐variable adjustment for race, education, and age at baseline was applied.ResultsOver a median follow up of 5.2 years, a total of 325 probable dementia and 640 mild cognitive impairment events occurred. After multi‐variable adjustment (as above), the hazard ratio per 10 mmHg decrease in systolic BP was 0.89 (95% confidence interval [CI] 0.79, 0.99) for probable dementia, 0.89 (95% CI 0.82, 0.96) for mild cognitive impairment, and 0.90 (95% CI 0.84, 0.97) for their composite outcome (Table). Dynamic predictions after 2 years of follow‐up provide a descriptive summary of systolic BP levels and predicted risk of probable dementia or mild cognitive impairment (Figure).ConclusionReducing systolic BP decreases the risk of both probable dementia and mild cognitive impairment.

Bayesian joint modeling of patient-reported longitudinal data on frequency and duration of migraine

In this methodological study, we address the joint modeling of longitudinal data on the frequency and duration migraine attacks collected from patients in a clinical study in which patients were repeatedly asked at each hospital visit to report the number of days of migraine attacks they had in the last $30$ days and the corresponding average duration of attacks. In our motivating data set, the migraine frequency outcome is a count variable inflated at multiples of $5$ and $10$ days, whereas the migraine duration outcome is reported entirely in discrete hours, including $0$ for non-migraine days and inflated at multiples of $12$ hours. In our study, we propose a joint modeling approach that models each migraine outcome by a multiple inflated negative binomial model with random effects and assumes a bivariate normal distribution for the random effects. We estimate the model parameters under Bayesian inference. We examine the performance of the proposed joint model using a Monte Carlo simulation study and compare its performance with a separate modeling approach in which each longitudinal count outcome is modeled separately. Finally, we present the results of the analysis of migraine data.

Bayesian semiparametric joint modeling of a count outcome and inconveniently timed longitudinal predictors.

The Women's Health Initiative (WHI) Life and Longevity After Cancer (LILAC) study is an excellent resource for studying the quality of life following breast cancer treatment. At study entry, women were asked about new symptoms that appeared following their initial cancer treatment. In this article, we were interested in using regression modeling to estimate associations of clinical and lifestyle factors at cancer diagnosis (independent variables) with the number of new symptoms (dependent variable). Although clinical and lifestyle data were collected longitudinally, few measurements were obtained at diagnosis or at a consistent timepoint prior to diagnosis, which complicates the analysis. Furthermore, parametric count models, such as the Poisson and negative binomial, do not fit the symptom data well. Thus, motivated by the issues encountered in LILAC, we propose two Bayesian joint models for longitudinal data and a count outcome. Our two models differ according to the assumption on the outcome distribution: one uses a negative binomial (NB) distribution and the other a nonparametric rounded mixture of Gaussians (RMG). The mean of each count distribution is dependent on imputed values of continuous, binary, and ordinal variables at a time point of interest (e.g. diagnosis). To facilitate imputation, longitudinal variables are modeled jointly using a linear mixed model for a latent underlying normal random variable, and a Dirichlet process prior is assigned to the random subject-specific effects to relax distribution assumptions. In simulation studies, the RMG joint model exhibited superior power and predictive accuracy over the NB model when the data were not NB. The RMG joint model also outperformed an RMG model containing predictors imputed using the last value carried forward, which generated estimates that were biased toward the null. We used our models to examine the relationship between sleep health at diagnosis and the number of new symptoms following breast cancer treatment in LILAC.

Measuring self-regulation in everyday life: Reliability and validity of smartphone-based experiments in alcohol use disorder

Self-regulation, the ability to guide behavior according to one’s goals, plays an integral role in understanding loss of control over unwanted behaviors, for example in alcohol use disorder (AUD). Yet, experimental tasks that measure processes underlying self-regulation are not easy to deploy in contexts where such behaviors usually occur, namely outside the laboratory, and in clinical populations such as people with AUD. Moreover, lab-based tasks have been criticized for poor test–retest reliability and lack of construct validity. Smartphones can be used to deploy tasks in the field, but often require shorter versions of tasks, which may further decrease reliability. Here, we show that combining smartphone-based tasks with joint hierarchical modeling of longitudinal data can overcome at least some of these shortcomings. We test four short smartphone-based tasks outside the laboratory in a large sample (N = 488) of participants with AUD. Although task measures indeed have low reliability when data are analyzed traditionally by modeling each session separately, joint modeling of longitudinal data increases reliability to good and oftentimes excellent levels. We next test the measures’ construct validity and show that extracted latent factors are indeed in line with theoretical accounts of cognitive control and decision-making. Finally, we demonstrate that a resulting cognitive control factor relates to a real-life measure of drinking behavior and yields stronger correlations than single measures based on traditional analyses. Our findings demonstrate how short, smartphone-based task measures, when analyzed with joint hierarchical modeling and latent factor analysis, can overcome frequently reported shortcomings of experimental tasks.

Joint modeling of longitudinal data with informative cluster size adjusted for zero-inflation and a dependent terminal event.

Repeated measures are often collected in longitudinal follow-up from clinical trials and observational studies. In many situations, these measures are adherent to some specific event and are only available when it occurs; an example is serum creatinine from laboratory tests for hospitalized acute kidney injuries. The frequency of event recurrences is potentially correlated with overall health condition and hence may influence the distribution of the outcome measure of interest, leading to informative cluster size. In particular, there may be a large portion of subjects without any events, thus no longitudinal measures are available, which may be due to insusceptibility to such events or censoring before any events, and this zero-inflation nature of the data needs to be taken into account. On the other hand, there often exists a terminal event that may be correlated with the recurrent events. Previous work in this area suffered from the limitation that not all these issues were handled simultaneously. To address this deficiency, we propose a novel joint modeling approach for longitudinal data adjusting for zero-inflated and informative cluster size as well as a terminal event. A three-stage semiparametric likelihood-based approach is applied for parameter estimation and inference. Extensive simulations are conducted to evaluate the performance of our proposal. Finally, we utilize the Assessment, Serial Evaluation, and Subsequent Sequelae of Acute Kidney Injury (ASSESS-AKI) study for illustration.

Impact of Two Different Types of Rabbit ATG on Immune Reconstitution and Overall Results after Allogeneic Hematopoetic Stem Cell Transplantation for Acute Lymphoblastic Leukemia in Children

Introduction Thymoglobulin® and Grafalon® are 2 polyclonal rabbit anti-thymocyte globulin (ATG) used in allogeneic hematopoietic stem cell transplantation (allo-HSCT) to prevent graft rejection and graft-versus-host disease (GvHD). Differences in manufacturing lead to different types and concentrations of antibodies in each product. Few studies compare their effects on post-transplant immune reconstitution: indeed, the latter is essential to prevent post-transplant infections as well as relapse (in case of hematological malignancies). Methods We conducted a retrospective study on children or adolescents who received a first unrelated allo-HSCT for an Acute Lymphoblastic Leukemia (ALL) between 2007 and 2018, in the department of pediatric hematology of the Robert Debré Hospital in Paris, France. During this time period, 2 types of ATG were used: Thymoglobulin® at 7.5mg/kg or Grafalon® at 60mg/kg. We included patients with a B or T phenotype ALL aged 2 to 18 years at the time of HSCT indication. Stem cell source could be from unrelated matched or mismatched bone marrow, peripheral blood or cord blood. All patients received a myeloablative conditioning regimen consisting of 12 grays total body irradiation and etoposide (60mg/kg). We compared patients for engraftment, counts of total lymphocytes, B and T lymphocytes, T CD4+ and CD8+ lymphocytes and their sub-types (naïve, effector memory, central memory and effector memory RA+), regulatory T cells, NK cells, and compared lymphocyte proliferation assays at 1, 3, 6 and 12 months after transplant. We used joint models of longitudinal and survival data to further analyze evolution of total lymphocytes, B and T cells, T CD4+ and T CD8+ cells, and NK cells. Finally, we compared clinical outcomes between the 2 groups. Statistical analyses were performed according to EBMT guidelines (Iacobelli, 2013). Results 76 children were included: 23 received Thymoglobulin® and 53 received Grafalon®. Stem cell source was peripheral blood or bone marrow for 61 patients and cord blood for 15 patients. Initial characteristics of patients, disease status and HSCT were similar between the two groups except for the median total nucleated cells and CD34+ progenitor cells of the graft, that were significantly higher in the Grafalon® group. We found a delayed cumulative incidence of engraftment in patients who received Grafalon® compared to those who received Thymoglobulin®: 29 days [interquartile range 22-36] versus 22 days [19.5-29.5], p=0.004. We observed significantly higher counts of total and T CD4+ lymphocytes 1 month after transplant in patients who received Thymoglobulin® as compared to those who received Grafalon®: 238/mm3 [172-466] versus 137/mm3 [51-306] respectively (p=0.005), and 40/mm3 [31.5-70.5] versus 16/mm3 [2.5-41.5], respectively (p=0.011). No differences in immune cell population counts were found 3, 6 and 12 months after transplant (figure 1). However, joint models found no difference in the evolution of lymphocyte subsets analyzed depending on the ATG used, except a non-significant difference in slopes over time for CD8+ cells (p=0.13). In patients given Thymoglobulin® compared to those given Grafalon®, we found a higher proliferation in response to stimulation by CD3 3 month after HSCT (CD3 index 20.4 [12.4;41] versus 5.9 [3.3;6.4] respectively, p=0.0039), and a higher proliferation in response to stimulation by adenovirus (ADV) antigen 1 year after HSCT (ADV index 78.6 [58.8;163.1] versus 15.7 [2.8;59.3] respectively, p=0.027). We found no effect of ATG type on overall survival (OS), disease-free survival (DFS), treatment-related mortality and chronic GvHD. In a multivariate analysis, we observed no effect of the type of ATG on OS and DFS. Conclusions We found a higher count of total and T CD4+ lymphocytes 1 month after unrelated allo-HSCT for ALL in children who received Thymoglobulin® at 7.5mg/kg compared to those who received Grafalon® at 60mg/kg. From 3 months after HSCT, we found no difference in immune reconstitution. Joint models for longitudinal data and survival found no significant difference in evolution of the lymphocyte subsets analyzed. Clinical outcomes were similar between the two groups. However, we compared an intermediate dose of Thymoglobulin® to a high dose of Grafalon® that is not still the standard in clinical practice. A prospective study comparing Thymoglobulin® to a lower dose of Grafalon® would be needed. Disclosures Baruchel: Novartis: Consultancy, Honoraria; Servier: Consultancy, Honoraria; Bristol-Myers Squibb: Consultancy, Honoraria; Bellicum: Consultancy. Dalle:Novartis: Honoraria, Membership on an entity's Board of Directors or advisory committees; Sanofi-Genzyme: Consultancy, Honoraria, Membership on an entity's Board of Directors or advisory committees, Research Funding; Bellicum: Consultancy, Honoraria; Medac: Consultancy, Honoraria; Orchard: Consultancy, Honoraria, Membership on an entity's Board of Directors or advisory committees; Jazz Pharmaceuticals: Consultancy, Honoraria, Membership on an entity's Board of Directors or advisory committees; bluebird bio: Consultancy, Honoraria, Membership on an entity's Board of Directors or advisory committees; Incyte: Consultancy, Membership on an entity's Board of Directors or advisory committees; Gilead: Honoraria; AbbVie Pharmacyclics: Membership on an entity's Board of Directors or advisory committees.

Joint Modeling of a Longitudinal Measurement and Parametric Survival Data with Application to Primary Biliary Cirrhosis Study

Although longitudinal and survival data are collected in the same study, they are usually analyzed separately. Measurement errors and missing data problems arise because of separate analysis of these two data. Therefore, joint model should be used instead of separate analysis. The standard joint model frequently used in the literature is obtained by combining the linear mixed effect model of longitudinal data and Cox regression model with survival data. Nevertheless, to use the Cox regression model for survival data, the assumption of proportional hazards must be provided. Parametric survival sub-models should be used instead of the Cox regression model for the survival sub-model of the JM where the assumption is not provided. In this article, parametric joint modeling of longitudinal data and survival data that do not provide the assumption of proportional hazards are investigated. For the survival data, the model with Exponential, Weibull, Log-normal, Log-logistic, and Gamma accelerated failure time models and the linear mixed effect model are combined with random effects and the models were applied in primary biliary cirrhosis data set obtained from Mayo Clinic. After determining the best parametric joint model according to Akaike and Bayesian information criterions, the best available model was compared with standard joint model and of separate analysis of survival data and longitudinal data. As a results, in the studies where longitudinal and survival data are obtained together, it is seen that the parametric joint model gives more better results than the standard joint model when the proportional hazard assumption is not provided.

Serum albumin changes and mortality risk of peritoneal dialysis patients.

Initial or single time point serum albumin levels have been shown to be important in predicting the prognosis of peritoneal dialysis (PD) patients. However, we assume the dynamic change and trend of albumin after PD are essential. We aimed to investigate the association between baseline albumin levels, albumin trajectories, and patient mortality in a retrospective cohort study. In this retrospective cohort study, 547 incident PD patients were enrolled from Peking University Third Hospital. Date were collected by medical records review, including age, gender, body mass index, primary disease, comorbidities, and laboratory tests. A joint model for longitudinal data and time-to-event data was used to establish the relationship between serum albumin trajectories and mortality risk of PD patients. The albumin trajectories was negatively correlated with risk of death. The increase in the current value of albumin trajectories at time points t after PD was associated with decreased risk of death (HR = 0.881, p < 0.0001). There was no statistical association between initial albumin and risk of death (HR = 1.030, 95% CI 0. 995-1.066). The results showed that increased age, higher albumin-corrected Ca levels, and higher eGFR values were risk factors for death. In addition, predictors of low albumin levels are increased PD time, increased age, increased albumin-corrected Ca, and decreased BMI as well as initial albumin levels. This study demonstrates that albumin trajectories after PD is better than initial serum albumin level in predicting mortality risk. Increasing albumin level over time can improve the prognosis of PD patients.

Missingness Mechanism that Incorporated Joint Modeling of Longitudinal Data with Monotone Dropout

We analyzed repeated measurement of continuous responses with monotone dropout. We are interested in reducing the bias associated with treatment effects, but the results' credibility relies on the validity of the techniques applied to analyze the data, and under the conditions where the techniques gives reliable answers. Furthermore, the robustness of the trial findings are determined through the application of sensitivity analysis which verifies to which extent the results are affected by changes in techniques, values of unmeasured variables and model assumptions. Moreover, the results obtain from the missing not at random (MNAR) is the same as their counterpart in missing at random (MAR). In addition, using multiple imputation (MI) in the analysis also improves the accuracy of results.

Bayesian Joint Modelling of Longitudinal Data on Abstinence, Frequency and Intensity of Drinking in Alcoholism Trials

In alcoholism research, several complementary outcomes are of interest:abstinence from drinking during a specific time frame, and, when the individual is drinking, frequency of drinking (the proportion of days on which drinking occurs) and intensity of drinking (the average number of drinks per drinking day). The outcomes are often measured repeatedly over time on the same subject and, although they are closely related, they are rarely modelled together. We propose a joint model that allows us to fit these longitudinal outcomes simultaneously, using correlated random effects to model the association between the outcomes and between repeated measurements on the same subject. The model has three parts: a logistic part for sustained abstinence over the period of interest, a truncated binomial part for frequency of drinking and a log-normal model for drinking intensity when drinking occurs. Because of the computational impracticality of fitting models with many random effects by using standard frequentist approaches, we use a Bayesian approach to fit the joint model. We also conduct a simulation study to investigate the gains in parameter estimate bias and mean-squared error associated with joint versus separate modelling. We illustrate the approach on data from an alcoholism clinical trial.

Bayesian quantile regression-based partially linear mixed-effects joint models for longitudinal data with multiple features.

In longitudinal AIDS studies, it is of interest to investigate the relationship between HIV viral load and CD4 cell counts, as well as the complicated time effect. Most of common models to analyze such complex longitudinal data are based on mean-regression, which fails to provide efficient estimates due to outliers and/or heavy tails. Quantile regression-based partially linear mixed-effects models, a special case of semiparametric models enjoying benefits of both parametric and nonparametric models, have the flexibility to monitor the viral dynamics nonparametrically and detect the varying CD4 effects parametrically at different quantiles of viral load. Meanwhile, it is critical to consider various data features of repeated measurements, including left-censoring due to a limit of detection, covariate measurement error, and asymmetric distribution. In this research, we first establish a Bayesian joint models that accounts for all these data features simultaneously in the framework of quantile regression-based partially linear mixed-effects models. The proposed models are applied to analyze the Multicenter AIDS Cohort Study (MACS) data. Simulation studies are also conducted to assess the performance of the proposed methods under different scenarios.

Bayesian quantile regression for nonlinear mixed-effects joint models for longitudinal data in the presence of mismeasured covariate errors

ABSTRACTQuantile regression (QR) models have recently received increasing attention in longitudinal studies where measurements of the same individuals are taken repeatedly over time. When continuous (longitudinal) responses follow a distribution that is quite different from a normal distribution, usual mean regression (MR)-based linear models may fail to produce efficient estimators, whereas QR-based linear models may perform satisfactorily. To the best of our knowledge, there have been very few studies on QR-based nonlinear models for longitudinal data in comparison to MR-based nonlinear models. In this article, we study QR-based nonlinear mixed-effects (NLME) joint models for longitudinal data with non-central location and outliers and/or heavy tails in response, and non-normality and measurement errors in covariate under Bayesian framework. The proposed QR-based modeling method is compared with an MR-based one by an AIDS clinical dataset and through simulation studies. The proposed QR joint modeling approach can be not only applied to AIDS clinical studies, but also may have general applications in other fields as long as relevant technical specifications are met.

Tutorial in Joint Modeling and Prediction: A Statistical Software for Correlated Longitudinal Outcomes, Recurrent Events and a Terminal Event

Extensions in the field of joint modeling of correlated data and dynamic predictions improve the development of prognosis research. The R package frailtypack provides estimations of various joint models for longitudinal data and survival events. In particular, it fits models for recurrent events and a terminal event (frailtyPenal), models for two survival outcomes for clustered data (frailtyPenal), models for two types of recurrent events and a terminal event (multivPenal), models for a longitudinal biomarker and a terminal event (longiPenal) and models for a longitudinal biomarker, recurrent events and a terminal event (trivPenal). The estimators are obtained using a standard and penalized maximum likelihood approach, each model function allows to evaluate goodness-of-fit analyses and plots of baseline hazard functions. Finally, the package provides individual dynamic predictions of the terminal event and evaluation of predictive accuracy. This paper presents theoretical models with estimation techniques, applies the methods for predictions and illustrates frailtypack functions details with examples.

Simultaneous Bayesian inference on a finite mixture of mixed-effects Tobit joint models for longitudinal data with multiple features

Simultaneous Bayesian inference on a finite mixture of mixed-effects Tobit joint models for longitudinal data with multiple features

Penalized spline joint models for longitudinal and time-to-event data

ABSTRACTThe joint models for longitudinal data and time-to-event data have recently received numerous attention in clinical and epidemiologic studies. Our interest is in modeling the relationship between event time outcomes and internal time-dependent covariates. In practice, the longitudinal responses often show non linear and fluctuated curves. Therefore, the main aim of this paper is to use penalized splines with a truncated polynomial basis to parameterize the non linear longitudinal process. Then, the linear mixed-effects model is applied to subject-specific curves and to control the smoothing. The association between the dropout process and longitudinal outcomes is modeled through a proportional hazard model. Two types of baseline risk functions are considered, namely a Gompertz distribution and a piecewise constant model. The resulting models are referred to as penalized spline joint models; an extension of the standard joint models. The expectation conditional maximization (ECM) algorithm is applied to estimate the parameters in the proposed models. To validate the proposed algorithm, extensive simulation studies were implemented followed by a case study. In summary, the penalized spline joint models provide a new approach for joint models that have improved the existing standard joint models.

Joint modeling of longitudinal data and discrete-time survival outcome.

A predictive joint shared parameter model is proposed for discrete time-to-event and longitudinal data. A discrete survival model with frailty and a generalized linear mixed model for the longitudinal data are joined to predict the probability of events. This joint model focuses on predicting discrete time-to-event outcome, taking advantage of repeated measurements. We show that the probability of an event in a time window can be more precisely predicted by incorporating the longitudinal measurements. The model was investigated by comparison with a two-step model and a discrete-time survival model. Results from both a study on the occurrence of tuberculosis and simulated data show that the joint model is superior to the other models in discrimination ability, especially as the latent variables related to both survival times and the longitudinal measurements depart from 0.

Bayesian inference on partially linear mixed-effects joint models for longitudinal data with multiple features

The relationship between viral load and CD4 cell count is one of the interesting questions in AIDS research. Statistical models are powerful tools for clarifying this important problem. Partially linear mixed-effects (PLME) model which accounts for the unknown function of time effect is one of the important models for this purpose. Meanwhile, the mixed-effects modeling approach is suitable for the longitudinal data analysis. However, the complex process of data collection in clinical trials has made it impossible to rely on one particular model to address the issues. Asymmetric distribution, measurement error and left censoring are features commonly arisen in longitudinal studies. It is crucial to take into account these features in the modeling process to achieve reliable estimation and valid conclusion. In this article, we establish a joint model that accounts for all these features in the framework of PLME models. A Bayesian inferential procedure is proposed to estimate parameters in the joint model. A real data example is analyzed to demonstrate the proposed modeling approach for inference and the results are reported by comparing various scenarios-based models.

Joint modeling of longitudinal data with a dependent terminal event

ABSTRACTLongitudinal data often arise in longitudinal follow-up studies, and there may exist a dependent terminal event such as death that stops the follow-up. In this article, we propose a new joint modeling for the analysis of longitudinal data with informative observation times via a dependent terminal event and two latent variables. Estimating equations are developed for parameter estimation, and asymptotic properties of the resulting estimators are established. In addition, a generalization of the joint model with time-varying coefficients for the longitudinal response variable is considered, and goodness-of-fit methods for assessing the adequacy of the model are also provided. The proposed method works well in our simulation studies, and is applied to a data set from a bladder cancer study.

A mixture of hierarchical joint models for longitudinal data with heterogeneity, non-normality, missingness, and covariate measurement error

ABSTRACTLongitudinal data arise frequently in medical studies and it is a common practice to analyze such complex data with nonlinear mixed-effects (NLME) models. However, the following four issues may be critical in longitudinal data analysis. (i) A homogeneous population assumption for models may be unrealistically obscuring important features of between-subject and within-subject variations; (ii) normality assumption for model errors may not always give robust and reliable results, in particular, if the data exhibit skewness; (iii) the responses may be missing and the missingness may be nonignorable; and (iv) some covariates of interest may often be measured with substantial errors. When carrying out statistical inference in such settings, it is important to account for the effects of these data features; otherwise, erroneous or even misleading results may be produced. Inferential procedures can be complicated dramatically when these four data features arise. In this article, the Bayesian joint modeling approach based on a finite mixture of NLME joint models with skew distributions is developed to study simultaneous impact of these four data features, allowing estimates of both model parameters and class membership probabilities at population and individual levels. A real data example is analyzed to demonstrate the proposed methodologies, and to compare various scenarios-based potential models with different specifications of distributions.