Brown Bag Lunch Schedule: Difference between revisions

From hcil
Jump to navigation Jump to search
(sigfried)
No edit summary
Line 98: Line 98:
|     
|     
'''Sigfried Gold''',<br>University of Maryland, College Park
'''Sigfried Gold''',<br>University of Maryland, College Park
| <div class="mw-collapsible mw-collapsed">
| <div class="mw-collapsible mw-collapsed">
Important medical research is increasingly based on analysis of data collected during provision of routine care. Compared to clinical trials data, this "secondary use" data is not susceptible to randomized, prospective study protocols; it suffers from poor quality and extreme "missingness" for observational or retrospective methods; strict privacy and human subjects regulations limit its availability; processing it for analysis is complicated by the diversity of its sources, formats, and the plethora of language and coding systems in which it is recorded; and analyzing it generally requires advanced clinical training and methods for grappling with its extreme multi-variateness, sparsity, and unknown systemic biases. Despite these formidable challenges, this data is orders of magnitude cheaper and more prolific than clinical trial data. Researchers and analysts within medical provider institutions can have access to data for millions of patients essentially for free; while medical products companies, regulators, and payer institutions can affordably purchase data for hundreds of millions of patients. Further, although analysts' uses cases are diverse and their methods (e.g., advanced statistics or machine learning) often opaque as well as immature; they share many basic questions and tasks: they almost universally need to characterize their populations on various demographic and clinical dimensions; they generally need to choose study and comparator cohorts; they need to group patients by disease and treatment parameters; they need to evaluate the significance of untold co-morbidities and confounders; they need to explore and discover temporal patterns obscured by the volume and variability of the data.
'''Exploratory visualization tools for health records research, and an exciting detour into infrastructural support for health records research at UMD''
<div class="mw-collapsible-content">
'''Abstract:''' Important medical research is increasingly based on analysis of data collected during provision of routine care. Compared to clinical trials data, this "secondary use" data is not susceptible to randomized, prospective study protocols; it suffers from poor quality and extreme "missingness" for observational or retrospective methods; strict privacy and human subjects regulations limit its availability; processing it for analysis is complicated by the diversity of its sources, formats, and the plethora of language and coding systems in which it is recorded; and analyzing it generally requires advanced clinical training and methods for grappling with its extreme multi-variateness, sparsity, and unknown systemic biases. Despite these formidable challenges, this data is orders of magnitude cheaper and more prolific than clinical trial data. Researchers and analysts within medical provider institutions can have access to data for millions of patients essentially for free; while medical products companies, regulators, and payer institutions can affordably purchase data for hundreds of millions of patients. Further, although analysts' uses cases are diverse and their methods (e.g., advanced statistics or machine learning) often opaque as well as immature; they share many basic questions and tasks: they almost universally need to characterize their populations on various demographic and clinical dimensions; they generally need to choose study and comparator cohorts; they need to group patients by disease and treatment parameters; they need to evaluate the significance of untold co-morbidities and confounders; they need to explore and discover temporal patterns obscured by the volume and variability of the data.


The advent of common data models and open-source software is just beginning to drastically streamline research workflows with this data. For analysts with access to data in OHDSI (ohdsi.org) format, for instance, many months of the standard observational study workflow can be skipped entirely. OHDSI's web-based cohort construction tools and it's open and growing R methods library allow researchers not only to define and execute their studies in hours or days rather than months, these researchers can now instantly and precisely share their code and aggregate results in a research network to be immediately replicated on dozens of other databases containing records for hundreds of millions of patients.
The advent of common data models and open-source software is just beginning to drastically streamline research workflows with this data. For analysts with access to data in OHDSI (ohdsi.org) format, for instance, many months of the standard observational study workflow can be skipped entirely. OHDSI's web-based cohort construction tools and it's open and growing R methods library allow researchers not only to define and execute their studies in hours or days rather than months, these researchers can now instantly and precisely share their code and aggregate results in a research network to be immediately replicated on dozens of other databases containing records for hundreds of millions of patients.
Line 108: Line 110:


At the Brown Bag I will talk about both; but depending on audience interest (most of our visualization researchers will be off at IEEE VIS this week), I may end up focusing more on the infrastructural issues.
At the Brown Bag I will talk about both; but depending on audience interest (most of our visualization researchers will be off at IEEE VIS this week), I may end up focusing more on the infrastructural issues.
<br><br>
In this talk I'll address how advances in machine learning can ameliorate these issues. Specifically, my students and I will introduce ongoing work on three problems: 1) how to use data to understand and simplify complex, high-dimensional, design spaces (to aid in techniques like optimization, design synthesis, and design exploration), 2) how to filter high-quality, diverse submissions out of large pools of design ideas generated by online communities (to aid in design generation and selection), and 3) how to enable non-experts to design complex mechanical parts (such as 3D printable robots) by using AI to automate various mechanical design tasks. Each problem highlights how building probabilistic models of designs via data can often produce a whole that is greater than the sum of its parts and make design (even of complex, physical systems) more inclusive.
<br><br>
<br><br>
'''Bio:''' With 29 years of experience in developing data management and analysis software on Unix/Linux and web platforms, Gold specializes in designing and implementing innovative, browser-based information visual analytics tools to facilitate the exploration and understanding of complex, multivariate or temporal data. He has experience in a wide array of industries (cyber security, securities trading, law, public sector administration, fundraising), but particular expertise in medical informatics and the secondary use of clinical and claims data for pharmacoepidemiology and patient safety research. He works with medical data using a common data model and open-source software as a collaborator in the OHDSI community.  
'''Bio:''' With 29 years of experience in developing data management and analysis software on Unix/Linux and web platforms, Gold specializes in designing and implementing innovative, browser-based information visual analytics tools to facilitate the exploration and understanding of complex, multivariate or temporal data. He has experience in a wide array of industries (cyber security, securities trading, law, public sector administration, fundraising), but particular expertise in medical informatics and the secondary use of clinical and claims data for pharmacoepidemiology and patient safety research. He works with medical data using a common data model and open-source software as a collaborator in the OHDSI community.  

Revision as of 13:30, 2 October 2017

The HCIL has an open, semi-organized weekly "brown bag lunch (BBL)" every Thursdays from 12:30-1:30pm in HCIL (2105 Hornbake, South Wing). The topics range from someone's work, current interests in the HCIL, software demos/reviews, study design, proposed research topics, introductions to new people, etc. The BBL is the one hour a week where we all come together--thus, it’s a unique time for HCIL members with unique opportunities to help build collaborations, increase awareness of each other’s activities, and generally just have a bit of fun together. There is no RSVP; simply show up!

If you would like to give or suggest a talk, presentation, workshop, etc., send an email to BBL student co-coordinators Sriram Karthik Badam (sbadam@umd.edu) or Pavithra Ramasamy (pavithra.ramasamy94@gmail.com). In the email, briefly describe the topic and preferred dates.

To be notified about upcoming events, please subscribe one of these mailing lists.



Fall 2017 Schedule

Date Leader Topic
08/31/2017

Kickoff to a new Semester!

Come network, make introductions, and share what each of us is working on

09/07/2017

David Weintrop, University of Maryland, College Park

To block or not to block: Understanding the effects of programming language representation in high school computer science classrooms.

09/14/2017

Stacy Branham,
University of Maryland Baltimore-County

From Independence to Interdependence: A Social Narrative of Assistive Technology

09/21/2017

Gabriela Marcu,
Drexel University
Cody Buntain,
University of Maryland, College Park

Gabriela: Addressing health inequities through human-centered design
Cody: Gaining Insight into Real-World Societal Response Using Social Media

09/28/2017

Mark Fuge,
University of Maryland, College Park

Designing with Data: How machine learning is morphing human, product, and system design

10/05/2017

Sigfried Gold,
University of Maryland, College Park

'Exploratory visualization tools for health records research, and an exciting detour into infrastructural support for health records research at UMD

10/12/2017

Foad Hamidi,
University of Maryland, Baltimore County

TBD

10/19/2017

Internship Panel? (TBD)

TBD

10/26/2017

Janet Walkoe,
University of Maryland, College Park

Technologically Mediated Teacher Noticing

11/02/2017

Hernisa Kacorri,
University of Maryland, College Park

TBD

11/09/2017

Karen Holtzblatt and Chris Robeck
University of Maryland, College Park

TBD

11/16/2017

Karthik Ramani,
Purdue University,
West Lafayette

TBD

11/23/2017 No Brown Bag, Thanksgiving recess
11/30/2017

Georgia Bullen

TBD

12/07/2017

Pamela Wisniewski
University of Central Florida

TBD

Past Brown Bags

View the Past Brown Bag Lunch Schedules to learn more about prior talks.