Projects by Researcher

Scientific Approaches

We plan to primarily work with EHR data contained in All of Us for a cohort of adult participants diagnosed with primary open-angle glaucoma. We will extract data on systemic conditions and medications for this cohort, as well as physical measurements and vital signs. We will clean the data such that the format is consistent with the data from our previously published model. Then, we will use this data as an external validation of a logistic regression model derived from our prior study that was based at a single academic center. Next, we will use All of Us data to train a new set of models, using techniques such as logistic regression, random forests, and artificial neural networks. We will optimize these models using feature selection methods and class balancing procedures. By evaluating performance metrics such as area under the curve (AUC), precision, recall, and accuracy, we will assess whether we can achieve superior predictive performance when training models using All of Us.

Scientific Approaches

Arrays will be imputed using the phasing and imputation tools Eagle2 and Minimac4. Polygenic risk score will then be calculated using the population genomics tool PLINK. A simple linear model will then be fit to the scores, which attempt to describe the macroscopic relationship between genetic ancestry and observed polygenic scores. The fitted parameters of this model can then be used to reduce genetic ancestry-dependent bias when calculating these scores in a clinical setting.

Scientific Approaches

We plan to employ a blood typing algorithm to extensively type RBC antigens from 1) whole genome sequencing and 2) array data in the AllofUs cohort, and compare the two outcomes. Then, we plan to employ the phenome-wide association study (PheWAS) approach to identify associations between RBC antigen types and other clinical phenotypes. PheWAS will be carried out using multivariable linear regression and logistic regressions with ABO blood groups with our novel ABO blood type. For example, in the case of the ABO blood group, ABO blood subtypes (A101, A102, Aw01, B101, etc.) will act as the independent variable and phenotypes, derived from participant provided information (PPI) electronic health records (EHR), as the dependent variable. Initial models will include adjustments for age, gender, and race/ethnicity. Differential associations by race/ethnicity, gender, and sex will also be evaluated.

Scientific Approaches

By running the notebooks in this workspace, you should get familiar with how to query PPI questions/surveys, what the frequencies of answers for each question in each PPI module are.

Scientific Approaches

We will annotate genomic variants from AoU participants with variant curations that have been recorded by the HGSC-CL clinical annotation team in its ‘VIP’ database or in ClinVar. We will assess the frequency of these previously-known pathogenic mutations, and provide breakdowns by ancestry.

Scientific Approaches

To characterize the diversity of the All of Us cohort, we analyzed participant genetic, demographic, and geographic data.

Here is a brief list of methods used:

1. All of Us participant genome-wide genotype was merged and harmonized with global reference population data.

2. Unsupervised clustering analysis techniques - Hopkins statistic, visual assessment of clustering tendency, K-means clustering & UMAP - to assess the extent of genetic structure in the cohort.

3. Supervised genetic ancestry inference using global reference populations, principal components analysis, and the Rye (Rapid ancestrY Estimation) program.

4. Genetic ancestry was compared to participants' self-identified race & ethnicity.

5. Geocoded data and participant age were used to measure how genetic ancestry and admixture vary with respect to participant geography and age.

6. Admixture regression to associate participant health outcomes, gleaned from electronic health records, with their genetic ancestry.

Scientific Approaches

Not applicable - these notebooks demonstrate example analysis how to use Hail and PLINK to perform genome-wide association studies using the All of Us genomic data and phenotypic data.

Scientific Approaches

N/A

Scientific Approaches

This Tutorial Workspace contains two Jupyter Notebooks (one written in Python, the other in R). Each notebook is divided into the following sections:

1. Setup: How to set up this notebook, install and import software packages, and select the correct version of the CDR.
2. Data Availability Part 1: How to summarize the number of unique participants with major data types: Physical Measurements, Survey, and EHR;
3. Data Availability Part 2: How to delve a little deeper into data availability within each major data type;
4. Data Organization: An explanation of how data is organized according to our common data model.
5. Example Queries: How to directly query the CDR, using two examples of SQL queries to extract demographic data.
6. Expert Tip: How to access the base version of the CDR, for users that want to do their own cleaning.

Scientific Approaches

We will select all All of Us participants who self-reported sex at birth male or female, whose self-reported race was white, black or Asian, as well as those who self-reported being Hispanics.

Atrial fibrillation (AF) will be identified from self-reports in the medical survey or from electronic health records (EHR).

Clinical factors will be identified from EHR and study measurements (blood pressure, weight, height).

We will evaluate the association of demographic (age, sex, race/ethnicity) and clinical (body mass index, blood pressure, smoking, cardiovascular diseases) factors with prevalence of self-reported AF and prevalence of AF in the EHR, as well as incident AF ascertained from the EHR.

Scientific Approaches

We will identify prevalence of opioid use in two ways and stratified by state.
First, we use EHR Drug Exposures to capture use of prescription opioid.
Second, we use lifestyle survey questionnaire to capture substance use reported by patients themselves:
1. In your LIFETIME, which of the following substances have you ever used?
2. In the PAST THREE MONTHS, how often have you used this substance?
The prevalence will be stratified by state, therefore EHR Observation Table will be used to capture this information.

Scientific Approaches

In this cross-sectional, population-based study, we used All of Us baseline data from patient (age>18) provided information (PPI) surveys and electronic health record (EHR) blood pressure measurements and retrospectively examined the prevalence of hypertension in the EHR cohort using Systemized Nomenclature of Medicine (SNOMED codes and blood pressure medications recorded in the EHR. We used the EHR data (SNOMED codes on 2 distinct dates and at least one hypertension medication) as the primary definition, and then add subjects with elevated systolic or elevated diastolic blood pressure on measurements 2 and 3 from PPI. We extracted each participant’s detailed dates of SNOMED code for essential hypertension from the Researcher Workbench table ‘cb_search_all_events’. We calculated an age-standardized HTN prevalence according to the age distribution of the U.S. Census, using 3 groups (18-39, 40-59, ≥ 60).