3 July 2014 in Journal of the American Statistical Association
We propose a Bayesian generalized low-rank regression model (GLRR) for the analysis of both high-dimensional responses and covariates. This development is motivated by performing searches for associations between genetic variants and brain imaging phenotypes. GLRR integrates a low rank matrix to approximate the high-dimensional regression coefficient matrix of GLRR and a dynamic factor model to model the high-dimensional covariance matrix of brain imaging phenotypes. Local hypothesis testing is developed to identify significant covariates on high-dimensional responses. Posterior computation proceeds via an efficient Markov chain Monte Carlo algorithm. A simulation study is performed to evaluate the finite sample performance of GLRR and its comparison with several competing approaches. We apply GLRR to investigate the impact of 1071 SNPs on top 40 genes reported by AlzGene database on the volumes of 93 regions of interest (ROI) obtained from Alzheimer’s Disease Neuroimaging Initiative (ADNI). Supplementary materials for this article are available online.
13 February 2016 in Cerebral Cortex
Brain structural covariance networks (SCNs) composed of regions with correlated variation are altered in neuropsychiatric disease and change with age. Little is known about the development of SCNs in early childhood, a period of rapid cortical growth. We investigated the development of structural and maturational covariance networks, including default, dorsal attention, primary visual and sensorimotor networks in a longitudinal population of 118 children after birth to 2 years old and compared them with intrinsic functional connectivity networks. We found that structural covariance of all networks exhibit strong correlations mostly limited to their seed regions. By Age 2, default and dorsal attention structural networks are much less distributed compared with their functional maps. The maturational covariance maps, however, revealed significant couplings in rates of change between distributed regions, which partially recapitulate their functional networks. The structural and maturational covariance of the primary visual and sensorimotor networks shows similar patterns to the corresponding functional networks. Results indicate that functional networks are in place prior to structural networks, that correlated structural patterns in adult may arise in part from coordinated cortical maturation, and that regional co-activation in functional networks may guide and refine the maturation of SCNs over childhood development.
26 October 2016 in Cerebral Cortex
Individual differences in neuroanatomy are associated with intellectual ability and psychiatric risk. Factors responsible for this variability remain poorly understood. We tested whether 17 major demographic and obstetric variables were associated with individual differences in brain volumes in 756 neonates assessed with MRI. Gestational age at MRI, sex, gestational age at birth, and birthweight were the most significant predictors, explaining 31% to 59% of variance. Unexpectedly, earlier born babies had larger brains than later born babies after adjusting for other predictors. Our results suggest earlier born children experience accelerated brain growth, either as a consequence of the richer sensory environment they experience outside the womb or in response to other factors associated with delivery. In the full sample, maternal and paternal education, maternal ethnicity, maternal smoking, and maternal psychiatric history showed marginal associations with brain volumes, whereas maternal age, paternal age, paternal ethnicity, paternal psychiatric history, and income did not. Effects of parental education and maternal ethnicity are partially mediated by differences in birthweight. Remaining effects may reflect differences in genetic variation or cultural capital. In particular late initiation of prenatal care could negatively impact brain development. Findings could inform public health policy aimed at optimizing child development.
1 September 2015 in NeuroImage
10 July 2021 in Journal of Clinical Oncology
PURPOSE: Brentuximab vedotin, an effective anti-CD30 antibody-drug conjugate approved for use in adults with classical Hodgkin lymphoma (HL), was introduced in this frontline trial to reduce prescribed radiation in children and adolescents with classical HL. METHODS: Open-label, single-arm, multicenter trial for patients (age ≤ 18 years) with stage IIB, IIIB, or IV classical HL was conducted. Brentuximab vedotin replaced each vincristine in the OEPA/COPDac (vincristine, etoposide, prednisone, and doxorubicin/cyclophosphamide, vincristine, prednisone, and dacarbazine) regimen according to GPOH-HD2002 treatment group 3 (TG3); two cycles of AEPA and four cycles of CAPDac. Residual node radiotherapy (25.5 Gy) was given at the end of all chemotherapy only to nodal sites that did not achieve a complete response (CR) at the early response assessment (ERA) after two cycles of therapy. Primary objectives were to evaluate the safety and efficacy (complete remission at ERA) of this combination and the 3-year event-free (EFS) and overall survival (OS). The trials are registered at ClinicalTrials.gov (identifier: NCT01920932 ). RESULTS: Of the 77 patients enrolled in the study, 27 (35%) achieved complete remission at ERA and were spared radiation. Patients who were irradiated received radiation to individual residual nodal tissue. At a median follow-up of 3.4 years, the 3-year EFS was 97.4% (SE 2.3%) and the OS was 98.7% (SE 1.6%). One irradiated patient experienced disease progression at the end of therapy and now remains disease free more than 6 years following salvage therapy, and one unexpected death occurred. Only 4% of patients experienced grade 3 neuropathy. CONCLUSION: The integration of brentuximab vedotin in the frontline treatment of pediatric high-risk HL is highly tolerable, facilitated significant reduction in radiation exposure, and yielded excellent outcomes.
17 June 2016 in Multivariate Behavioral Research
Factor analysis is a popular statistical technique for multivariate data analysis. Developments in the structural equation modeling framework have enabled the use of hybrid confirmatory/exploratory approaches in which factor-loading structures can be explored relatively flexibly within a confirmatory factor analysis (CFA) framework. Recently, Muthén & Asparouhov proposed a Bayesian structural equation modeling (BSEM) approach to explore the presence of cross loadings in CFA models. We show that the issue of determining factor-loading patterns may be formulated as a Bayesian variable selection problem in which Muthén and Asparouhov's approach can be regarded as a BSEM approach with ridge regression prior (BSEM-RP). We propose another Bayesian approach, denoted herein as the Bayesian structural equation modeling with spike-and-slab prior (BSEM-SSP), which serves as a one-stage alternative to the BSEM-RP. We review the theoretical advantages and disadvantages of both approaches and compare their empirical performance relative to two modification indices-based approaches and exploratory factor analysis with target rotation. A teacher stress scale data set is used to demonstrate our approach.