Estimating Multilevel Models on Data Streams.

Social scientists are often faced with data that have a nested structure: pupils are nested within schools, employees are nested within companies, or repeated measurements are nested within individuals. Nested data are typically analyzed using multilevel models. However, when data sets are extremely large or when new data continuously augment the data set, estimating multilevel models can be challenging: the current algorithms used to fit multilevel models repeatedly revisit all data points and end up consuming much time and computer memory. This is especially troublesome when predictions are needed in real time and observations keep streaming in. We address this problem by introducing the Streaming Expectation Maximization Approximation (SEMA) algorithm for fitting multilevel models online (or "row-by-row"). In an extensive simulation study, we demonstrate the performance of SEMA compared to traditional methods of fitting multilevel models. Next, SEMA is used to analyze an empirical data stream. The accuracy of SEMA is competitive to current state-of-the-art methods while being orders of magnitude faster.

Casting wider nets for anxiety and depression: disability-driven cross-diagnostic subtypes in a large cohort.

BACKGROUND: In search of empirical classifications of depression and anxiety, most subtyping studies focus solely on symptoms and do so within a single disorder. This study aimed to identify and validate cross-diagnostic subtypes by simultaneously considering symptoms of depression and anxiety, and disability measures. METHOD: A large cohort of adults (Lifelines, n = 73 403) had a full assessment of 16 symptoms of mood and anxiety disorders, and measurement of physical, social and occupational disability. The best-fitting subtyping model was identified by comparing different hybrid mixture models with and without disability covariates on fit criteria in an independent test sample. The best model's classes were compared across a range of external variables. RESULTS: The best-fitting Mixed Measurement Item Response Theory model with disability covariates identified five classes. Accounting for disability improved differentiation between people reporting isolated non-specific symptoms ['Somatic' (13.0%), and 'Worried' (14.0%)] and psychopathological symptoms ['Subclinical' (8.8%), and 'Clinical' (3.3%)]. Classes showed distinct associations with clinically relevant external variables [e.g. somatization: odds ratio (OR) 8.1-12.3, and chronic stress: OR 3.7-4.4]. The Subclinical class reported symptomatology at subthreshold levels while experiencing disability. No pure depression or anxiety, but only mixed classes were found. CONCLUSIONS: An empirical classification model, incorporating both symptoms and disability identified clearly distinct cross-diagnostic subtypes, indicating that diagnostic nets should be cast wider than current phenomenology-based categorical systems.

Model-based approaches to synthesize microarray data: a unifying review using mixture of SEMs.

Several statistical methods are nowadays available for the analysis of gene expression data recorded through microarray technology. In this article, we take a closer look at several Gaussian mixture models which have recently been proposed to model gene expression data. It can be shown that these are special cases of a more general model, called the mixture of structural equation models (mixture of SEMs), which has been developed in psychometrics. This model combines mixture modelling and SEMs by assuming that component-specific means and variances are subject to a SEM. The connection with SEM is useful for at least two reasons: (1) it shows the basic assumptions of existing methods more explicitly and (2) it helps in straightforward development of alternative mixture models for gene expression data with alternative mean/covariance structures. Different specifications of mixture of SEMs for clustering gene expression data are illustrated using two benchmark datasets.

A mixture model with random-effects components for classifying sibling pairs.

In healthy aging research, typically multiple health outcomes are measured, representing health status. The aim of this paper was to develop a model-based clustering approach to identify homogeneous sibling pairs according to their health status. Model-based clustering approaches will be considered on the basis of linear mixed effect model for the mixture components. Class memberships of siblings within pairs are allowed to be correlated, and within a class the correlation between siblings is modeled using random sibling pair effects. We propose an expectation-maximization algorithm for maximum likelihood estimation. Model performance is evaluated via simulations in terms of estimating the correct parameters, degree of agreement, and the ability to detect the correct number of clusters. The performance of our model is compared with the performance of standard model-based clustering approaches. The methods are used to classify sibling pairs from the Leiden Longevity Study according to their health status. Our results suggest that homogeneous healthy sibling pairs are associated with a longer life span. Software is available for fitting the new models.

Fetal growth trajectories in Type-1 diabetic pregnancy.

OBJECTIVE: To describe the individual intrauterine growth patterns of fetuses of insulin-dependent (Type-1) diabetic women and to examine determinants of overgrowth (macrosomia) and its timing. METHODS: This retrospective longitudinal study examined the developmental trajectories of fetal abdominal circumference (AC) and biparietal diameter in 76 Type-1 diabetic women with singleton pregnancies. Latent class analysis was used to identify subgroups of patients with a shared fetal AC growth trajectory. Subsequently, maternal factors, including glycemic control as assessed by glycosylated hemoglobin (HbA1c), were examined to see whether they had any effect on fetal growth. RESULTS: Four subgroups with different AC growth patterns were identified. Differences in birth weight between the distinct subgroups were related to the shape of the AC growth velocity curve over gestation. Acceleration of AC growth commencing before or after 25 weeks' gestation was associated with the birth of a heavy or large-for-dates baby in 94 and 56% of cases, respectively. Poor glycemic control (HbA1c > 7.0%) during the periconception period or before 12 weeks' gestation was a modest predictor of midtrimester growth in AC. Other diabetes-related factors, fetal sex, parity, or maternal weight/obesity were unrelated to the fetal growth pattern. CONCLUSION: The findings suggest that an individual fetus's growth trajectory is set early in gestation and that the contemporaneous degree of maternal glycemia plays a role in determining birth weight.

Beyond symptom dimensions: schizophrenia risk factors for patient groups derived by latent class analysis.

INTRODUCTION: Patients grouped by latent class analysis of symptoms show some consensus between studies, and may be less etiologically heterogeneous than current diagnoses. If so, the effect size of 'neurodevelopmental' risk factors may be greater than in equivalent DSMIV diagnostic groups. METHOD: Two hundred fifty six individuals with neurodevelopmental risk factors recorded in the National Child Development Study (1958) UK birth cohort were grouped by data-driven illness subtypes, derived previously in over 1000 individuals. The effect sizes of these risks were compared between data-derived and DSMIV schizophrenia (295.x) groups. RESULTS: Compared to DSMIV schizophrenia, the data-driven subtype broadly characterized by the presence of psychotic symptoms in the absence of affective symptoms showed significantly greater effect sizes in eight out of thirteen continuously-rated risk factors: birth weight, cognition, childhood behavioural problems, and neurological softsigns including handedness. CONCLUSION: A data-driven subgroup of schizophrenia patients, characterized as lacking co-morbid depressive symptoms, is less heterogeneous with respect to neurodevelopmental etiology.

Incidence and completeness of notification of Legionnaires' disease in The Netherlands: covariate capture-recapture analysis acknowledging regional differences.

To estimate incidence and completeness of notification of Legionnaires' disease (LD) in The Netherlands in 2000 and 2001, we performed a capture-recapture analysis using three registers: Notifications, Laboratory results and Hospital admissions. After record-linkage, 373 of the 780 LD patients identified were notified. Ascertained under-notification was 52.2%. Because of expected and observed regional differences in the incidence rate of LD, alternatively to conventional log-linear capture-recapture models, a covariate (region) capture-recapture model, not previously used for estimating infectious disease incidence, was specified and estimated 886 LD patients (95% confidence interval 827-1022). Estimated under-notification was 57.9%. Notified, ascertained and estimated average annual incidence rates of LD were 1.15, 2.42 and 2.77/100 000 inhabitants respectively, with the highest incidence in the southern region of The Netherlands. Covariate capture-recapture analysis acknowledging regional differences of LD incidence appears to reduce bias in the estimated national incidence rate.

[Births: developments in the past and expectations for the future]. / Geboorte: ontwikkelingen in het verleden en toekomstverwachtingen.

PIP: Revisions in the fertility assumptions made in the 1991 national population forecasts for the Netherlands are described. The changes reflect the continuing increase in fertility rates of women aged over 30, which have led to a reduction in the estimates of women remaining childless in the projections. The author notes the growing assumption that female labor force participation can be effectively combined with having and raising children. (SUMMARY IN ENG)^ieng

[Life-style and demographic behavior: an application of latent class analysis]. / Leefstijl en demografisch gedrag: een toepassing van latente-klasse-analyse.

"The aim of this study is to construct a lifestyle typology which is relevant for demographic behaviour. The concept lifestyle is defined as attitudes towards marriage, family, parenthood and the labour force participation of women. The lifestyle indicators available in the 1988 Netherlands Fertility Survey are used to construct a typology by means of latent-class-models.... The lifestyle typology can be used in two ways for projection purposes. Firstly, it can be as an explanatory variable in models for demographic behaviour. This application is demonstrated with a discrete-time hazard model for first births....Secondly, it can be used to identify innovators with respect to demographic behaviour and to quantify alternative hypotheses about the number of followers. This is demonstrated by calculating the future proportion of childlessness." (SUMMARY IN ENG)

[Young and single in the 1980s and 1990s]. / Jong en alleenstaand in de jaren '80 en '90.

"In the Netherlands increasing numbers of young people who leave their parents' home choose to live alone. This behavioural change dates back to the 60s....[and] is based on general social changes which are not restricted to the Netherlands.... Based on the Netherlands Fertility Survey '88...[a model is developed that] indicates that emancipation-indicators strongly affect the choice of young women to live alone. These are: participation in education, educational attainment and participation in the labour market. The number of young women living alone in the year 2000 has been calculated on the basis of forecasts of the education- and labour-market participation...." (SUMMARY IN ENG)