We foster dialogue between data scientists and researchers in clinics and laboratories in order to drive excellence in health care research at Stanford.

About the Data Studio

The Data Studio is a collaboration between Spectrum (The Stanford Center for Clinical and Translational Research and Education) and the Department of Biomedical Data Science. The Data Studio is open to the Stanford community engaged in biomedical research. We expect it to have educational value for students and postdocs interested in biomedical data science. The Data Studio features DBDS faculty and staff who offer the following services: workshops, office hours, and one-to-one consultations. When you complete the Data Studio request form, our coordinator and consultants will work with you to choose the right service for your research project. Appointments may be requested by completing the required form.

Workshops are an extensive and in-depth consultation for a Medical School researcher based on research questions, data, statistical models, and other material prepared by the researcher with the aid of our facilitator. During the Data Studio Workshop, the researcher explains the project, goals, and needs. Experts in the related topic from across campus will be invited and contribute to the brainstorming. After the meeting, the facilitator will follow up, helping with immediate action items and summary of the discussion. Ultimately, we strive to pair each PI with a data scientist for long-term collaboration.Office Hours are brief consultations for Medical School researchers during the last session of each month. DBDS faculty are available to advise about your research questions. Consult the schedule below to complete the Office Hour registration form. Once you have registered, you will receive a calendar invitation with the date, time, and location of the session. Bring any data, prior analyses, or other materials that you have. Our consultants may even recommend your project for a Workshop if it is appropriate.

One-to-one consultations for Medical School researchers are available year-round. Our facilitator assigns each request to a data scientist with the relevant expertise.

Partners

General questions about statistical issues may be brought to the STAT390 Consulting Workshop. This is a class offered by the Department of Statistics during each academic quarter that is staffed by graduate students and directed by a faculty instructor. The service typically consists of a single meeting with the researcher to address a specific concern, such as planning of experiments and data analysis. For more information, consult the STAT390 Consulting Workshop web page.

Researchers who are members of the Stanford Cancer Institute (SCI) conducting research projects related to cancer may request assistance from the SCI Biostatistics Shared Resource.

Schedule

The Data Studio is held each Wednesday from 3:00 until 4:30 pm during the fall, winter, and spring quarters of the academic year. Consult the schedule below for the location of each session. Students may participate by enrolling in BIODS 232 for an introduction to the art of statistical consultation and practicum working on projects with a biomedical researcher. All are welcome to attend. Click here to sign up for our mailing list.

Currently scheduled topics are listed below, followed by links to topics and summaries from previous quarters.

Cardiac Events after Radiation of Chemotherapy in Breast Cancer Patients

DATE: Wednesday, 29 November 2023

TIME: 3:00–4:30 PM

LOCATION: Conference Room X399, Medical School Office Building, 1265 Welch Road, Stanford, CA

INVESTIGATORS:

Scott Jackson (1)

Michael Binkley (1)

Department of Radiation Oncology

WEBPAGE: https://dbds.stanford.edu/data-studio/

ABSTRACT

The Data Studio Workshop brings together a biomedical investigator with a group of experts for an in-depth session to solicit advice about statistical and study design issues that arise while planning or conducting a research project. This week, the investigator(s) will discuss the following project with the group.

INTRODUCTION

This observational study consists of two patient groups: the treatment group receives combined radiation with Chemotherapy (XRT+Chemo) and the control group receives Chemotherapy (Chemo). Our project concerns competing risk regression for cardiac events. Death is a competing risk. Some of the covariates of interest only apply to the XRT+Chemo patients, namely, those related to radiation.

HYPOTHESIS & AIM

What is the risk for breast cancer patients of cardiac events after either XRT+Chemo or Chemo?

DATASET

The dataset includes 228 breast cancer patients treated at Stanford with either XRT+Chemo (treatment, n=174) or Chemo (controls, n=54).

STATISTICAL MODELS

We have fitted the following models to the dataset:

Kaplan-Meier curves for comparing overall survival between Chemo and XRT+Chemo cohorts
Fine-Gray curves for comparing cumulative incidence of cardiac events (using first event if multiple) with death as a competing risk.
A set of univariable competing risk regression models was also utilized for a large set of covariates, with group (Chemo or XRT+Chemo) considered as a clustering variable.

ADJUSTMENT FOR BIAS

For these radiation covariates (collectively called X), just setting the value to zero for X in the control arm may not necessarily be a good approach. If X predicts the survival response for the treatment arm (suppose that is the case and that is why you want to include them in the covariates), you may want to see if other covariates available in both arms can predict X in the treatment arm. If they can, you may want to impute the missing X in the control arm to reduce the bias. An example was a recently finished trial in which patient satisfaction with the tool immediately after the intervention is highly predictive of the final outcomes. Of course, satisfaction with the tool can be collected only in the intervention arm and itself is a trial outcome (i.e., mediator). Several baseline covariates could predict patient satisfaction with the tool in the treatment arm. The investigators used a predictive X using the equation derived from treatment arm and applied Xhat to both arms and found interactions between Xhat and treatment arms in one of the two co-primary endpoints. It turned out certain patients were more likely to satisfy the treatment and often had better response than those unlikely to satisfy the treatment. We just want to show how to impute the unmeasured covariates depends on the nature of this study.

STATISTICAL QUESTIONS

The question concerns competing risks in time-to-event analysis with covariates that are not measurable in one arm.

Is the use of a cluster term appropriate for treatment?
1. How does one decide between cluster or frailty in these models, when we assume outcomes within a given cohort are correlated in some way?
2. I tried using both cluster() and frailty() and the results were similar.
3. Therneau and Grambsch discuss the differences between cluster() and frailty() as follows in Modeling Survival Data, Extending the Cox Model (pp. 169-170):

Random effects or frailty model, such as that described in Oakes [119]. The model includes a random per-subject effect; multiple outcomes are assumed to be independent conditional on the per-subject coefficient.
A marginal models approach. This has much in common with the generalized estimating equations (GEE) approach of Zeger, et al. [168]. (Referring to cluster)

It was suggested to set radiation-specific variables (which are all numeric) to 0 for Chemo patients.
1. Is this appropriate?
2. Why or why not?

ZOOM MEETING INFORMATION

Join from PC, Mac, Linux, iOS or Android: https://stanford.zoom.us/j/96972699747?pwd=RVh5NjBPdzFvcjdNWHQ5cCtQQnVNdz09

Password: 842586

Or iPhone one-tap (US Toll): +18333021536,,96972699747# or +16507249799,,96972699747#

Or Telephone:

Dial: +1 650 724 9799 (US, Canada, Caribbean Toll) or +1 833 302 1536 (US, Canada, Caribbean Toll Free)

Meeting ID: 969 7269 9747

Password: 842586

International numbers available: https://stanford.zoom.us/u/adNUBM1izD

Meeting ID: 969 7269 9747

Password: 842586

SIP: 96972699747@zoomcrc.com

Password: 842586

Data Studio Office Hour

DATE: Wednesday, 15 November 2023

TIME: 3:00–4:30 PM

LOCATION: Conference Room X399, Medical School Office Building, 1265 Welch Road, Stanford, CA

REGISTRATION FORM: https://redcap.stanford.edu/surveys/?s=WMH74XCX33

DESCRIPTION:

The Data Studio Office Hour brings together a series of biomedical investigators with a group of experts for brief individualized sessions to solicit advice about a statistical and study design issue that arises while planning or conducting a research project.

This week, Data Studio holds office hours for your data science needs. Biomedical Data Science faculty are available to provide assistance with your research questions. If you need help with bioinformatics software and pipelines, check out the Computational Services and Bioinformatics Facility (http://cmgm-new.stanford.edu/) and the Genetics Bioinformatics Service Center (http://med.stanford.edu/gbsc.html).

Reserve a Data Studio Office Hour session by completing the Registration Form. Sessions are about 30 minutes long but might be extended at the discretion of the coordinator. If you register for a session, please be present at the start time on Wednesday.

If you are not able to register for a session, you are welcome to complete our Data Studio Consultation services form for a free one-hour meeting with one of our statisticians. You will find a link to the Consultation services form on our Data Studio web page (https://dbds.stanford.edu/data-studio/).

ZOOM MEETING INFORMATION:

Join from PC, Mac, Linux, iOS or Android: https://stanford.zoom.us/j/96972699747?pwd=RVh5NjBPdzFvcjdNWHQ5cCtQQnVNdz09

Password: 842586

Or iPhone one-tap (US Toll): +18333021536,,96972699747# or +16507249799,,96972699747#

Or Telephone:

Dial: +1 650 724 9799 (US, Canada, Caribbean Toll) or +1 833 302 1536 (US, Canada, Caribbean Toll Free)
Meeting ID: 969 7269 9747
Password: 842586
International numbers available: https://stanford.zoom.us/u/adNUBM1izD

Meeting ID: 969 7269 9747

Password: 842586

SIP: 96972699747@zoomcrc.com

Password: 842586

Data Studio Office Hour

DATE: Wednesday, 1 November 2023
TIME: 3:00–4:30 PM

LOCATION: Conference Room X399, Medical School Office Building, 1265 Welch Road, Stanford, CA
REGISTRATION FORM: https://redcap.stanford.edu/surveys/?s=WMH74XCX33

DESCRIPTION

ZOOM MEETING INFORMATION:

Join from PC, Mac, Linux, iOS or Android: https://stanford.zoom.us/j/96972699747?pwd=RVh5NjBPdzFvcjdNWHQ5cCtQQnVNdz09

Password: 842586

Or iPhone one-tap (US Toll): +18333021536,,96972699747# or +16507249799,,96972699747#

Or Telephone:

Dial: +1 650 724 9799 (US, Canada, Caribbean Toll) or +1 833 302 1536 (US, Canada, Caribbean Toll Free)

Meeting ID: 969 7269 9747
Password: 842586
International numbers available: https://stanford.zoom.us/u/adNUBM1izD
Meeting ID: 969 7269 9747
Password: 842586
SIP: 96972699747@zoomcrc.com

Password: 842586

Red Teaming Generative AI in Healthcare

DATE: Wednesday, 25 October 2023

TIME: 3:00–6:00 PM (Students may leave at 4:20 PM but are welcome to stay.)

LOCATION: Room E241, CHEM-H Building (Bldg. ID 14-220), 290 Jane Stanford Way, Stanford, CA (See attached PDF with a map of building and location of room.)

WEBPAGE: https://dbds.stanford.edu/event/workshop-on-genai-in-healthcare-red-teaming/

ABSTRACT

Our BIODS 232 (Data Studio) Workshop on Wednesday, 25 October, will be a joint session with the Biomedical Data Science Generative AI Workshop. Enrolled students must attend the workshop in person during the regularly scheduled class time (3:00 to 4:20 PM) but are welcome to stay for the entire workshop.

DESCRIPTION

Please join us for an interactive session to test GenAI models for potential issues with biases, inaccuracies, and other matters related to healthcare. Red-teaming is a form of evaluation that elicits model vulnerabilities that might lead to undesirable behaviors. We will be looking for such behaviors on healthcare-related tasks. We will break into groups with table leaders from Stanford Healthcare, School of Medicine, CS and external partners providing technical and clinical insights. At the end we’ll have a read-out of findings and then continue the conversations during happy hour.

SCHEDULE

3:00pm | Check in begins

3:15pm | Table groups begin Red Teaming

4:45pm | Happy hour

6:00pm | Event concludes

__________________________

Data Studio Office Hour

DATE: Wednesday, 18 October 2023

TIME: 3:00–4:30 PM

LOCATION: Conference Room X399, Medical School Office Building, 1265 Welch Road, Stanford, CA

REGISTRATION FORM: https://redcap.stanford.edu/surveys/?s=WMH74XCX33

DESCRIPTION

ZOOM MEETING INFORMATION:

Join from PC, Mac, Linux, iOS or Android: https://stanford.zoom.us/j/96972699747?pwd=RVh5NjBPdzFvcjdNWHQ5cCtQQnVNdz09

Password: 842586

Or iPhone one-tap (US Toll): +18333021536,,96972699747# or +16507249799,,96972699747#

Or Telephone:

Dial: +1 650 724 9799 (US, Canada, Caribbean Toll) or +1 833 302 1536 (US, Canada, Caribbean Toll Free)

Meeting ID: 969 7269 9747

Password: 842586

International numbers available: https://stanford.zoom.us/u/adNUBM1izD

Meeting ID: 969 7269 9747

Password: 842586

SIP: 96972699747@zoomcrc.com

Password: 842586

DATE: Wednesday, 11 October 2023

TIME: 3:00–4:30 PM

TITLE: A Novel Chest Computed Tomography Scoring System in Children and Adolescents with Rheumatologic Diffuse Lung Disease

LOCATION: Conference Room X399, Medical School Office Building, 1265 Welch Road, Stanford, CA

INVESTIGATORS:

Beverley Newman (1)

Michal Cidon (2)

Terry Robinson (3, Presenter)

Paul Iskander (4)

Paul J. Thacker (5)

Evan Zucker (1)

Tzielan Lee (6)

Lu Tian (7)

Rex Moats (3)

(1) Pediatric Radiology, Stanford Children’s Hospital

(2) Pediatric Rheumatology, Children’s Hospital, Los Angeles

(3) Pediatric Radiology, Children’s Hospital, Los Angeles

(4) Pediatric Radiology, UCLA Mattel Children’s Hospital, Los Angeles

(5) Pediatric Radiology, Mayo Clinic, Rochester, Minnesota

(6) Pediatric Rheumatology, Stanford Children’s Hospital

(7) Biomedical Data Science, Stanford University School of Medicine

WEBPAGE: http://med.stanford.edu/dbds/resources/data-studio.html

ABSTRACT

INTRODUCTION

Pulmonary involvement in pediatric rheumatologic disease is associated with long-term high morbidity and mortality, yet current management is challenging because there are no current standardized approaches to defining, diagnosing, and providing risk stratification in assessing the extent and severity of diffuse lung disease in pediatric rheumatology patients with pulmonary compromise. In addition, while chest CT findings can predict the extent and severity of lung disease in pediatric rheumatology patients earlier than pulmonary function tests and 6-minute walk tests which are usually abnormal when patients become symptomatic, there are no current clinical approaches that systematically evaluate these patients with clinically acceptable chest CT imaging protocols.

There is currently limited chest CT imaging information in pediatric rheumatology patients, based on either clinical case reports or limited chest CT scoring primarily in adults and one study in children who were assessed for systemic sclerosis.

HYPOTHESIS & AIM

Development of a pediatric rheumatologic chest CT scoring system will represent a major advance in the understanding of early and progressive pulmonary findings in children with specific rheumatologic diseases that are associated with pulmonary complications. We hypothesize that a pediatric-specific rheumatologic scoring system for diffuse lung disease, will identify lung parenchymal findings associated with early and progressive lung disease. The purpose of this project was to establish a robust chest CT scoring system using inter-observer and intra-observer agreement of several chest CT imaging markers of diffuse lung disease to detect active and chronic progressive pulmonary changes in pediatric patients with different rheumatologic conditions. To develop such a scoring system, a previous adult Systemic Sclerosis Interstitial Lung Disease Scoring system developed by Golden and associates (Chest, 2007) was modified and adapted for a variety of lung conditions associated with different pediatric rheumatologic diseases.

We anticipate this pediatric diffuse lung disease score will be used to monitor treatment effects as well as being used as a clinical validation tool for further quantitative chest CT imaging analysis using currently developing deep learning algorithms for chest CT evaluation of diffuse interstitial lung disease in pediatric patients.

DATASET

We have curated one hundred twenty (120) chest CT scans obtained over a 10-year period (2011–2020). Three groups are included: (A) Disease Group (DG, N=42) of pediatric rheumatology patients with defined pulmonary involvement; (B) Disease Control Group (DCG, N=34) of pediatric rheumatology patients with no defined pulmonary disease; and (C) Normal Control Group (NCG, N=44) of age-matched normal control subjects.

METHODS

This study is the largest imaging comparison of DG, DCG, and NCG. The chest CT scans in our dataset were initially scored by three (3) board-certified pediatric radiologists with Certificates of Added Qualifications (CAQ) and 5–10 years of experience reading thoracic chest CT scans. All scoring was done after initial training with a pediatric thoracic radiologist with greater than 30 years of experience reading thoracic chest CT scans and pediatric pulmonologist with greater than 25 years of experience evaluating chest CT imaging, who developed the Pediatric Rheumatologic Diffuse Lung Disease Score [PRDLDS]. All pediatric radiology scoring was done by the 3 pediatric radiologists who were blinded to the clinical cases, clinical disease, and duration of findings.

After initial chest CT scoring evaluation in the N=120 scans by the three (3) pediatric radiology scorers, the PRDLD score was modified (PRDLD Score à mPRDLD score) to address inconsistent findings due to technical limitations of the majority of chest CT scans (non-contrast chest CT scans) or no specific adult findings that were not seen in pediatric patients. To further evaluate inter-reader reliability, after more extensive training using the mPRDLD score, the same 3 pediatric radiologists re-scored N=60 scans (DG: N=30; DCG: N=20; NCG: N=10) of the original N=120 scans 10 months later. To further address intra-reader reliability, N=30 scans (DG: N=20; DCG: N=10; NCG: N=5) of the 60 scans were further scored 4 months after completion of the N=60 scoring. In each re-scoring assessment, all de-identified scans were again randomized prior to reader evaluation and scoring.

In collaboration with Imbio, Inc., a CT image processing company, N=95 (N=42 DG subjects; N=33 DCG subjects; & N=20 NCG subjects) of the N=120 cases have been further utilized for a joint CHLA-Imbio, Inc. NIH SBIR Phase I research project for developing a deep learning (neural network) application for quantitative pediatric interstitial lung disease evaluation (Lung DeepLTA). We plan to compare the clinical scoring results from the mPRDLD Scores with that of the quantitative scores generated by Imbio Lung DeepLTA. In addition, we also have an internal funded CHLA research project evaluating standard lower dose chest CT and ultra-low dose chest CT in pediatric rheumatology patients at CHLA for the development of ultra-low dose chest CT imaging for serial pediatric rheumatologic assessment of early and progressive pulmonary disease in these patients.

GOALS

We hope to validate the current Pediatric Rheumatologic Diffuse Lung Disease Score. We have utilized an initial scoring assessment of subcomponent scores and modified scoring components based on initial scoring findings. Utilizing our inter-rater reliability and intra-rater reliability assessments, we hope to validate this new scoring system. Ultimately, we are interested in comparing the currently developing deep learning quantitative chest CT measurements (Lung DeepLTA) with the clinical imaging scoring system (mPRDLDS) we have developed for pediatric rheumatologic diffuse lung disease.

STATISTICAL QUESTIONS

(1) What is the best method for validating our clinical scoring system (mPRDLDS)? What are the best statistics we can use for validating a new clinical scoring system for a manuscript we are submitting to Pediatric Rheumatology? Given no gold standard, we will need good internal validation of the scoring system, and a means to address both accuracy and precision.

(2) Given limitations in the small numbers of subjects in certain pediatric rheumatology disease groups, what statistics can we use?

(3) We have a scoring system that provides regional assessment of disease extent/severity in six (6) lobar regions. What are the best statistics to use for demonstrating regional findings for each type of pediatric rheumatology disease? (Upper lobe vs. lower lobe predominance) differences in lobe distribution: RUL, RML, RLL, LUL, Lingula, LLL.

(4) What are the best statistical methods to compare clinical chest CT scores versus Lung DeepLTA findings? How can we best develop statistical analysis for DeepLTA findings for the different pediatric rheumatologic diseases that have similar or different lung diseases?

ZOOM MEETING INFORMATION

Join from PC, Mac, Linux, iOS or Android: https://stanford.zoom.us/j/96972699747?pwd=RVh5NjBPdzFvcjdNWHQ5cCtQQnVNdz09

Password: 842586

Or iPhone one-tap (US Toll): +18333021536,,96972699747# or +16507249799,,96972699747#

Or Telephone:

Dial: +1 650 724 9799 (US, Canada, Caribbean Toll) or +1 833 302 1536 (US, Canada, Caribbean Toll Free)

Meeting ID: 969 7269 9747

Password: 842586

International numbers available: https://stanford.zoom.us/u/adNUBM1izD

Meeting ID: 969 7269 9747

Password: 842586

SIP: 96972699747@zoomcrc.com

TITLE: Clinical Trial Design for Glaucoma Treatment Using Humphrey Visual Field as Primary Outcome

INVESTIGATORS:
Laurel Stell, Biomedical Data Science
Jeffrey Goldberg, Ophthalmology
Gala Beykin, Ophthalmology

DATE: Wednesday, 4 October 2023
TIME: 3:00–4:30 PM
LOCATION: Conference Room X399, Medical School Office Building, 1265 Welch Road, Stanford, CA

ABSTRACT
The Data Studio Workshop brings together a biomedical investigator with a group of experts for an in-depth session to solicit advice about statistical and study design issues that arise while planning or conducting a research project. This week, the investigator(s) will discuss the following project with the group.

Introduction
Glaucoma treatments are typically assessed by whether they control interocular pressure (IOP), but the disease often continues to progress despite reduction in IOP. The Humphrey Visual Field (HVF) exam, which measures the retina’s sensitivity to light, is widely used to diagnose glaucoma and its progress, but its measurement error can be large in comparison to the rate of progression. Consequently, estimating the rate of decrease in HVF measurements by linear regression generally requires regular exams over 10 years or more, and even then the slope is often not statistically significant. Finally, treatments are not likely to reverse damage but only slow or delay neurodegeneration. All of these factors can result in prohibitively large sample sizes or long trial times when using HVF as primary outcome in a clinical trial.
Hypothesis & Aim
We have performed exploratory analysis of HVF exams. We hope to leverage such data to improve clinical trial inclusion criteria and statistical tests for treatment effect.
Dataset
We have HVF data from a variety of sources: (a) thirty glaucomatous eyes in a test-retest study that performed weekly exams for three months (Artes et al, 2014), (b) data from Phase 1b trials including six or fewer exams over a year or two from about 150 eyes (Goldberg et al, 2022), and (c) the public UW-HVF data set of thousands of eyes, including 450 with at least nine exams over 10 years or more–but without clinical information such as diagnosis, progression or treatment.
Statistical Models
The HVF exam measures sensitivity at an array of 52 points on the retina. We will discuss properties of the measurements at individual locations, averaged over the whole retina, and averaged over each of six regions identified by mapping neurons in the retina. We are seeking advice on statistical models for testing treatment effect.

STATISTICAL QUESTIONS:

Do we have sufficient pilot data?
If not, what do we need?
How to estimate power for possible outcome measures?ZOOM MEETING INFORMATION

Join from PC, Mac, Linux, iOS or Android: https://stanford.zoom.us/j/96972699747?pwd=RVh5NjBPdzFvcjdNWHQ5cCtQQnVNdz09
Password: 842586

Or iPhone one-tap (US Toll): +18333021536,,96972699747# or +16507249799,,96972699747#

Or Telephone:
Dial: +1 650 724 9799 (US, Canada, Caribbean Toll) or +1 833 302 1536 (US, Canada, Caribbean Toll Free)

Meeting ID: 969 7269 9747
Password: 842586
International numbers available: https://stanford.zoom.us/u/adNUBM1izD

Meeting ID: 969 7269 9747
Password: 842586

SIP: 96972699747@zoomcrc.com
Password: 842586

TITLE: A Guide for the Statistically Perplexed
DATE: Wednesday, 27 September 2023
TIME: 3:00–4:30 PM
LOCATION: Conference Room X399, Medical School Office Building, 1265 Welch Road, Stanford, CA

INSTRUCTORS:

Ying Lu
Chiara Sabatti
Lu Tian
Balasubramanian Narasimhan (Naras)
Brad Efron
Mei-Chiung Shih
John S. Tamaresis

WEBPAGE: https://dbds.stanford.edu/data-studio/

ABSTRACT:

This will be the first meeting of the Autumn Quarter. Our agenda will include: