Post on 24-Apr-2015
description
Sustainability & Cochrane ReviewsHow technology can help
Chris MavergamesDirector of Web DevelopmentThe Cochrane Collaboration
Structure of this talk•Sustainability of Review production
(internal sustainability)▫Technologies to assist Review production▫PICOtron and Overviews of Reviews
•Sustainability in the health information marketplace (external sustainability)▫“Nimble” content and thinking outside
Review “container”▫Linked data and Star Trek
•Summary
Information Technology Strategy
“Information and Communication Technology (ICT) is at the heart of The Cochrane Collaboration ... To a very large extent, the success of The Cochrane Collaboration has been based on its investment in ICT.”
- From the Collaboration’s information technology strategy paper (ISSF)
Our data and systems are great, but...
Can we make the machines work a bit
harder?
Some critical points
•For internal processes▫How to automate w/o creating more work▫Important that methodology remains sound
•For external consumers▫Retain context of findings and results▫Create better understanding, not confusion
Sustainability of Review production
(and updating)
Producing Reviews - challenges•We have this incredibly complex,
methodologically rigorous process and aim to be global and comprehensive with a mostly-volunteer workforce
•We’ve just cleared 5,000 reviews and it's taken nearly 20 years, and these need to be maintained and updated
• Estimates say we need min. 10,000 Reviews to be comprehensive (estimate 10 years old!)
Cochrane Reviews are fantastic but...
•…creating them is a long and laborious process
Cochrane Reviews are fantastic but...
•…creating them is a long and laborious process
Help!
Review production assistance• Reference managers
▫ EndNote, Zotero, Connotea, Mendeley and others
• Screening/appraising references▫ ScreenToGo App (email: c.huckvale@imperial.ac.uk for public beta)
• Automated abstract screening/appraisal▫ Semi-automated appraisal Support Vector Machines (SVMs)
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2824679/
• Software covering multiple steps in systematic review production▫ EPPI-Reviewer (http://eppi.ioe.ac.uk/cms/Default.aspx?tabid=1913)▫ Distiller SR (http://systematic-review.net/)▫ EROS (Early Review Organizing Software http://rs.iecs.org.ar/)▫ Automated data extraction from Cochrane Reviews (reading the XML)
• GRADEpro• CRS
Slide courtesy: Rachel Marshall
What if authors could...?
• Pull in Risk of Bias assessment information already done on trials or studies
• See studies already included or excluded in other Reviews
• Other links between studies and Reviews via the CRS – more later in Star Trek section
• Use datasets like Drugbank and Diseasome for auto-completion or help filling out fields in RevMan for standardisation
Could we use technology to assist with…?
•Using MeSH mappings to show coverage of Reviews
•Using study links to show where gaps exist
•Priority-setting
•Creating derivative “views” or products from Cochrane data
PICOtron
PICOtron and Cochrane Clinical Answers
•Project to semi-automate populating Cochrane Clinical Answers, a derivative product of The Cochrane Library
•Python script written by Iain Marshall that extracts data from Cochrane Reviews
•Script automatically and randomly reconfigures Review titles into one of 30 question formats
PICOtron•Data is pulled from various areas of the Review
XML
•Depending on whether the result is significant and favored intervention, a sentence (narrative result per PICO) explaining these results is automatically generated
• Then, an author checks these and combines this information across PICOs into a “clinical answer“
PICOtron output
Overviews of Reviews & Network Meta-analysis
Help in preparing of Overviews of Reviews and Network Meta-analysis driven by linked data
Image from Lorne Becker
paroxetine
sertraline
citalopram
escitalopram
fluoxetine
fluvoxamine
milnacipran
venlafaxine
reboxetine
bupropion
mirtazapine
duloxetine
12 new generation
antidepressants:
Which ones are the
most efficacious?
Example of a trial level synthesis
Thanks to Georgia Salanti. Source: http://cmimg.cochrane.org/workshops-19th-cochrane-colloquium-october-2011
paroxetine
sertraline
citalopram
fluoxetine
fluvoxamine
milnacipran
venlafaxine
reboxetine
bupropion
mirtazapine
duloxetine
escitalopram
Network of Randomized Trials
Thanks to Georgia Salanti. Source: http://cmimg.cochrane.org/workshops-19th-cochrane-colloquium-october-2011
paroxetine
sertraline
citalopram
fluoxetine
fluvoxamine
milnacipran
venlafaxine
reboxetine
bupropion
mirtazapine
duloxetine
escitalopram
Network of Randomized Trials
No trials comparing
reboxetine to bupropion
available in Cochrane Reviews
?
Thanks to Georgia Salanti. Source: http://cmimg.cochrane.org/workshops-19th-cochrane-colloquium-october-2011
Other ideas?
Perhaps we can gain insight into how to improve Review production processes from how our users actually use our output?
segue ...
Sustainability in the health information
marketplace
“The problem is not information overload but filter failure.”
– Clay Shirky
•Cochrane provides a vital curation, filtering and evaluation/quality-assessment role and we need to make this clear
Again, Cochrane Reviews are fantastic BUT...
•There are problems that limit their use by some people▫Difficult to wade through all of the text▫Difficult to understand the figures,
terminology, and other bits of the Review▫Hard to compare interventions without
reading multiple Reviews▫Moving from studies in CENTRAL to Reviews
that included that study difficult▫Can be difficult to find the Review you seek
Searching The Cochrane Library
•Search for “Prozac” – no reviews•Search for “fluoxetine” – 27 reviews
Ideally, we’d restructure our content for different users
•Beginning to do this now:▫Summaries.Cochrane.org for consumers▫Cochrane Clinical Answers for clinicians
•BUT▫Takes a lot of work to reformulate reviews &
authors, CRGs, etc are busy
Wouldn’t it be nice if we could automate or semi-automate some of these processes?
Thinking outside Review container
Thinking outside Review container
Making our content “nimble”
Structured and linked data can help make our content “nimble”
Nimble content can:
• Travel Freely• Retain Context Meaning• Create New Products
- R. Lovinger, Razorfish
Linked data
What is linked data?
Semantic Web
is made up of:
Linked Data & Web of Data
Which all together comprise
Web 3.0
Current web = Web of documents
Docs are linked not data in docs
Machines aren‘t good at reading web pages
•Data on the web is meant for human consumption
•Machines need the data to be structured
•Once structured, information can be more easily shared within datasets and across web pages
It‘s about...
• Taking the complex relationships, interactions and dependencies in our data and modeling them in semantic web language and concepts for machine processing...
• So that we can do things like:▫Gain insights into our data▫Help with priority setting▫Repackage it for different users▫Later, perhaps machine can infer new
knowledge from our data and/or when our data is combined with other datasets
Cochrane Register of Studies
CRS and CENTRAL
• Lack of unique study IDs a real problem
• CRS solves this by providing a unique ID for all studies that can be referenced
• Better linking of data about trials and to Reviews▫ Example: Using forest plots to generate related
studies lists for CENTRAL
• Possibilities with linking to external sources such as PubMed
Insert witty Star Trek reference here!
Findings ontology
Image: Lorne Becker
Marshall-o-gram
Marshall-o-gram
Cochrane Review ontology
We can…•Ask questions that use data from several different
reviews
• Improve search
•Make it easier for people to find Cochrane Reviews
• Link data from studies and Reviews better
• Enhance the experience of our users by including data from other datasets
A question using multiple reviews
I’ve done a search for trials on a particular intervention for dementia.
I want to know which of the trials have been included in a Cochrane Review and a
summary of the risks of bias for the entire set of trials.
Links to the relevant Review for those trials that were included
Links to the relevant Review for those trials that were included
This study was one of 38 studies included in the Cochrane Review, <Title Here>.
Click here to see the full review
These figures summarize Risks of Bias from the trials included in the reviews in your search
RoB Summary for Cochrane Reviews on dementia
Make search work better
Enhancing the User Experience
You Say “Paracetamol”I Say “Acetaminophen”•Or, one could say any of these: Abenol (CA), Acephen, Anadin Paracetamol
(UK), Apo-Acetaminophen (CA), Aspirin Free Anacin, Atasol (CA), Calpol (UK), Cetaphen, Children's Tylenol Soft Chews, Disprol (UK), Exdol (CA), Feverall, Galpamol (UK), Genapap, Genebs, Infant's Pain Reliever, Mandanol (UK), Nortemp, Pain Eze, Panadol (UK), Robigesic (CA), Silapap, Tycolene, Tylenol 8 Hour, Tylenol, Tylenol Arthritis, Uni-Ace, Valorin
Cochrane leading in Web 3.0?
EbHC Semantic Platform
CDSR CRS/CENTRAL
DARE
HTAs
CMR
EbHC Semantic Platform
CDSR CRS/CENTRAL
DARE
HTAs
CMR
DrugBank
UMLS
Diseasome
SymptomOntology
* BBCHealth
Ontology
* Not yet created
Cochrane and EbHC ontology?
Will Cochrane have a bubble here someday?
Summary
• Technology IS at the heart of what we do• For both internal and external applications, we
can leverage these tools to further our mission
•Requires that we think differently about the “container“ of the Review
•Our data needs to become “nimble“ to meet future user needs
•We should proceed slowly, incrementally - What are the “quick wins“?
•Cochrane has the chance to lead in Web 3.0
Thank you