Exploring Existing Data Resources Pilot Awardees
Multidisciplinary Pediatric Diabetes Care
Team: Christine March, MD (PI), Ingrid Libman, MD, PhD, Elizabeth Miller, MD, PhD, Julie Donahue, PhD, Scott Rothenberger, PhD
Abstract: Diabetes is one of the most common chronic diseases of childhood, and the incidence of both type 1 (T1D) and type 2 (T2D) diabetes is increasing. Youth with diabetes are at risk for higher morbidity and mortality compared to healthy peers. The best predictor of long-term complications is cumulative glycemic exposure. By and large, most youth with diabetes fail to meet recommended glycemic targets, despite modern treatment advances, with clear inequities in outcomes by diagnosis type, race/ethnicity, and socioeconomic status. Contributing to this landscape is access to multidisciplinary care from diabetes providers, diabetes educators, dieticians, and behavioral health providers. Though single center reports have examined missed appointments with select aspects of the diabetes care team in both T1D and T2D, fidelity to the multidisciplinary care model as a whole has never been studied in a generalizable cohort. Using Optum’s Integrated database, a robust national dataset composed of claims and electronic health data, we aim to describe and compare patterns of multidisciplinary care use among youth with T1D and T2D. We will identify characteristics of youth who experience multidisciplinary care and those with gaps in care using multivariable regression modeling. Finally, we will explore the relationship between multidisciplinary care and two key outcomes, glycemic control (Hemoglobin A1c) and acute care utilization using similar analyses. Completion of these aims will help to characterize issues in health care utilization facing this population, the initial step to designing novel, community-partnered interventions to strengthen support systems and bridge gaps in care.
Urine metabolomics exploring existing data in the clinical toxicology laboratory
Team: Kenichi Tamama MD, PhD (PI), Dhivyaa Rajasundaram, PhD
Abstract: Substance abuse or drug addiction has been one of the significant health hazards worldwide. Drug screening is routinely used in the clinical management of drug addicts, but biomarkers of drug addiction itself or chronic drug exposure are largely unavailable to clinicians currently. Availability of such biomarkers should help clinicians identify the patients with a high risk of developing drug addiction and better predict the disease trajectories and responses of the patients to treatments. The overall objective of this proposal is to determine novel substance abuse-associated biomarker candidates by exploring existing unused data acquired in urine comprehensive drug screening (CDS) at the Clinical Toxicology Laboratory. More than 99% of the acquired raw mass spectral data in urine CDS are untouched and wasted. In this proposal, we will perform the following specific aims in this pilot study to determine substance abuse-associated novel biomarker candidates. Aim 1: Identify the metabolites that are significantly associated with drug addiction. Aim 2: Determine the differential metabolomics profiles of the patients receiving buprenorphine or methadone. Upon successful completion of the proposed research, we expect our contribution to determine urinary metabolites associated with drug addiction (Aim 1) and the future retention of the patients in the medicationassisted treatment (Aim 2). This contribution will be significant because the availability of such biomarkers should help clinicians predict the disease trajectories and responses of the patients to treatments, ultimately leading to increased social productivity and reduced future overdose events of the patients through better clinical management.
Radio-Omics and Deep Learning Analysis for Hepatocellular Carcinoma
Team: Michele Molinari, MD (PI), Shandong Wu, PhD, Marta Minervini, MD, Alessandro Furlan, MD, Nick D’Ardenne, MD, Christopher Buros, MD
Abstract: Adults (>18 years) treated for HCC with hepatic resections at the University of Pittsburgh Medical Center between January 1, 2011, and December 31st, 2020, will be candidates for this study. Patients will be excluded if they were treated with palliative surgeries, if they had diagnosis of mixed cholangio-HCC, or if they had previous hepatic resections for malignancies. The study population will be divided into two groups: patients with HCC in the settings of NAFLD and patients with HCC from other risk factors (e.g., viral hepatitis, (B or C), alcoholic cirrhosis, cryptogenic cirrhosis, others). Baseline demographic, clinical, pathological and radiological characteristics between the two groups will be analyzed. The most recent contrast enhanced computerized tomographic (CT) scans or magnetic resonance imaging (MRI) performed before the index operations will be retrieved and de-identified. Staging radiological studies will be analyzed by experienced body-imaging radiologists to determine the maximum size of the largest HCC, the total number of tumors, the presence of vascular tumor invasion, the presence of peri-hepatic lymphatic metastases. For each participant, radiomics analysis of the largest tumor and surrounding liver parenchyma will be performed and analyzed with the assistance of artificial intelligence and deep learning technologies. The primary aim of the study is to determine if there are radiological difference between patients with HCC and NAFLD versus patients with HCC and other predisposing factors.
A3ST: AI-based Automated Fidelity Assessment for Strategy Training in Inpatient Rehabilitation
Team Members: Yanshan Wang, PhD (PI), Leming Zhou, PhD, Elizabeth Skidmore, PhD, Alexandra Harper, PhD
Abstract: Stroke is a leading cause of disability in the world. Cognitive impairments occur in as many as half of adults who sustain a first-time stroke and are associated with significant long-term disability. Efficacious interventions that promote independence could significantly improve the quality of life for these individuals and reduce health care expenses. Strategy training, a multidisciplinary rehabilitation approach developed by the team, teaches skills that can be used to significantly reduce disability. Randomized controlled clinical trials have shown that strategy training is a more feasible and efficacious intervention to promote independence than the traditional rehabilitation approaches. To assess the implementation of strategy training and treatment integrity, the team has developed a standardized fidelity assessment approach. This approach requires a highly trained independent evaluator to use a validated fidelity checklist to assess instructional cues (i.e., guided and directed verbal cues and gestures) from video and audio recordings of rehabilitation sessions. This fidelity assessment approach is valid and feasible in single-site studies, but it is labor intensive, time-consuming, expensive, and not scalable as the trail is expanding to 30 inpatient rehabilitation facilities across the nation. To address the challenge of examining the fidelity of intervention delivery in large clinical trials, we plan to leverage advanced artificial intelligence (AI) techniques to automate the fidelity assessment approach, which have the potential to propel and translate to rehabilitation intervention practice and research forward in new directions previously untapped. The proposed project will be among the first to develop automated deep neural network-based AI methods for characterizing the delivery of clinical rehabilitation interventions.
Novel Methods for Single-Cell DNA Methylation Data
Team Members: Jiebiao Wang, PhD (PI), Manqi Cai, Brandon McKinney, MD, PhD, Chris McKennan, PhD
Abstract: In the last decade, single-cell multi-modal omics data has been explosively collected and thus was selected as Method of the Year 2019 by Nature Methods, after single-cell sequencing was selected as Method of the Year 2013. As funded by NIH, the BRAIN (Brain Research through Advancing Innovative Neurotechnologies) Initiative constructed the Cell Census Network (BICCN) to conduct a multi-modal cell census and atlas of the brain. The initial product of BICCN for millions of cells of the primary motor cortex is published in Nature 2021. The rapid development and continued expansion of sequencing technologies produced a huge amount of data. Over one thousand computational tools have been developed to analyze single-cell RNA-sequencing (scRNAseq) data. However, little attention has been paid to analyzing single-cell DNA methylation (scDNAm) data, which is key to regulating gene transcription and disease development. The scDNAm data are sparse and noisy, thus providing an unprecedented opportunity for developing novel statistical and computation methods. Unlike existing work that mainly focused on data integration and cell clustering, we propose statistical methods to analyze the regulatory mechanisms between single-cell epigenetics (DNAm, chromatin accessibility, and conformation) and transcriptomics data and detect multi-omics cell type marker genes. With robust cell type markers, we will build a cell type signature matrix for DNAm and deconvolve tissue-level DNAm to infer cellular fractions. This can enhance the cell-type-specific interpretation of bulk DNAm data and reduce the confounding in tissue-level DNAm analysis introduced by cellular heterogeneity.