Thursday, May 15, 2014

#Health2stat – all about data: experts from #nih talk about #health and #bigdata

WebMD reaches 174M unique visitors every month

.h1 Translational science – Rosemary Filiart

Advancing research to improve human health.
Integrating multiple disciplines.
Research and clinical data sets, applying appropriate privacy protections and using big data to reveal understanding.

Re-purposing data in a predictable and reproducible way and advancing towards personalized medicine and providing actionable and informed decisions at the point of care.

.h1 Repurposing yesterday’s data – Lisa Federer

Radical re-use an example is shipping containers becoming hotel rooms

.h2 Facilitating re-use

Translate expertise from analog to digital – look to librarians
1. Description: Standardized metadata eg. pubmed
2. Discoverability: data catalogs – these build on metadata
3. Dissemination: facilitating sharing while protecting privacy and intellectual property
4. Digital infrastructure: cyber infrastructure
5. Data Literacy: equipping people with tools and knowledge to be able to access data

How do we re-think the future of data. We don’t throw data away so how do we prepare it for future use.

.h1 using clinical data at NIH – BTRIS data mining – Jim Cimino

The National Institutes of Health consists of 27 institutes and centers, many of which conduct clinical research. Research data are collected in the NIH Clinical Center’s electronic health record (EHR) and institute and laboratory systems. The Biomedical Translational Research Information System (BTRIS) is a repository that collects data from these sources to provide unified tools to support researchers in the analysis of their data. BTRIS is available to any NIH researcher who wishes to obtain data for secondary analyses to reexamine old questions or ask new ones. Non-NIH researchers can collaborate with NIH researchers in the analysis of BTRIS data.

50 data sources – mostly live daily feeds
Half a million patients
140,000 clinical concepts

They have a de-identified data set that goes back to 1976 that can be queried.

The self service query tool is incredibly powerful.

.h1 discovering medical knowledge using BTRIS – Vojtech Huser

Vojtech works with BTRIS using R and a few other tools.
Meta map is an internal nih tool.

There are challenges in de-identifying data. Search and replace has to be very precise because there are numerous conditions that are named using people’s names. Eg. Removing Parkinson could also remove Parkinson’s disease references.

One interesting request: don’t take your EHR data to heaven – donate it to science.
This is something that consumer-mediated exchanges like could facilitate.

.h1 TB a world health problem – Stefan Jaeger

How do we detect TB in remote populations?

Some challenges: HIV populations have weakened immune systems making them susceptible to TB.
Some strains of TB are drug resistant.

USAID – AMPATH partnership working in Africa. Developing a portable X-ray machine that can be taken to villages to test people.

The next step is to use automatic image processing to identify TB in X-rays since there aren’t enough radiologists in Africa to review X-rays manually.

.h1 new computational tools and models for data mining – Jim DeLeo

John Von Neumann – “machines can think” 1955

Moved from what can the machine do to “what do we want the machine to do”

Jim’s area of expertise is computational intelligence.this subsumes artificial intelligence.

Machine Learning – just like humans by looking at lots of data and cluster and classify the data.

Jim’s latest focus is extreme multi-disciplinary teaming.
Every team member is passionate about the work. Short term. Clear deliverables.

.h1 Deep learning – new computational tools for biomedical learning – Jonathan Simon

Deep neural networks – a new technique for analyzing large volumes of data.
Machine learning by using existing data

Neural networks – simple inter-connected computational units. Modeled on the way the brain works.

Deep neural networks – based on neural networks and have very many hidden layers of computation. These emerged since 2006 as we gained new tools to analyze data and computational power became more affordable.

Biomedical deep learning is co-opting tools like image processing and adapting to medical applications. These tools are limited by the amount of data that is available. Fortunately more and more biomedical data is being made available online every year.

Now for the Q&A…

Mark Scrimshire
Health & Cloud Technology Consultant

email: mark@ekivemark.comStay up-to-date: Twitter @ekivemark

via WordPress