-
P ER SP EC T I V E
REPRODUC IB I L I TY
Dow
nloaded
A “datathon” model to supportcross-disciplinary
collaborationJerôme Aboab,1* Leo Anthony Celi,1 Peter Charlton,1
Mengling Feng,1
Mohammad Ghassemi,1 Dominic C. Marshall,1† Louis Mayaud,1
Tristan Naumann,1 Ned McCague,1 Kenneth E. Paik,1 Tom J.
Pollard,1
Matthieu Resche-Rigon,1 Justin D. Salciccioli,1 David J.
Stone2,3
In recent years, there has been a growing focus on the
unreliability of published bio-medical and clinical research. To
introduce effective new scientific contributors to theculture of
health care, we propose a “datathon” or “hackathon” model in which
partici-pants with disparate, but potentially synergistic and
complementary, knowledge andskills effectively combine to address
questions faced by clinicians. The continuous peerreview
intrinsically provided by follow-up datathons, which take up prior
uncompletedprojects, might produce more reliable research, either
by providing a different perspec-tive on the study design and
methodology or by replication of prior analyses.
by guest on Novem
ber 23, 2017http://stm
.sciencemag.org/
from
The research problems of health care extendbeyond clinical
medicine and cannot be solvedby physicians working in isolation.
Cliniciansare well aware of the uncertainties andinformation gaps
that permeate the practiceof medicine, which range from the trivial
butannoying to the significant and seemingly in-tractable.However,
busy clinicians typically donot have the time, energy, or training
to ad-dress questions they encounter in day-to-daypractice.
Instead, scientists, physician-scientists,and engineers carry out
biomedical researchdesigned to decipher clinical problems, butthese
efforts often are isolated from criticalclinical input. In
addition, recent accusationsagainst this research community include
theproduction of an overwhelming number ofpublished studies that
are irreproducible,underpowered, poorly designed, and other-wise
lack statistical rigor (1–5).
The typical clinician solves problemsby ap-plying basic
medical-science concepts and di-agnostic and treatment protocols
masteredduring medical training. In academic facilities,there are
physicians who see patients butspend most of their time working on
basic,translational, or clinical research. In fact,
somephysician-scientists—even those on medicalschool
faculties—rarely have exposure to clin-
1Massachusetts Institute of Technology (MIT) CriticalData, MIT,
Cambridge, MA 02139, USA. 2Departmentsof Anesthesiology and
Neurosurgery, University of Vir-ginia School of Medicine,
Charlottesville, VA 22908, USA.3Center for Wireless Health,
University of Virginia Schoolof Engineering and Applied Science,
Charlottesville, VA22904, USA.*All authors contributed equally to
this manuscript.†Corresponding author. E-mail:
[email protected]
ical medicine. The nonresearcher cliniciansseem to be getting
busier and busier with pa-tient care, but their input is critically
importantto identify relevantmedical questions andprob-lems and to
tackle them with a deep under-standing of the clinical context.
This must beaccomplished in an era in which progressivelylimited
resources are struggling to cope withincreasingly expensive health
care.
Over the past decade or so, data scientistsand engineers have
become increasinglydrawn to and involved with health care (5).This
interest has recently been accelerated bythe now near-universal
digitization of healthcare, which provides data scientists with a
toe-hold and a progressively important role in dai-ly clinical care
as well as biomedical research.Howcan this groupof diverse talents,
interests,and schedules be combined for productive col-laboration?
We propose an adaptation of thecomputer industry’s hackathon
model—anevent in which diverse professionals (entrepre-neurs,
software developers, designers, engineers)work together on software
development over ashort period of time. In the biomedical arena,
ahackathon similarly would bring together par-ticipants with
disparate but complementaryknowledge and skills—for example,
academicbiomedical researchers, industry scientists,bioengineers,
physicians, statisticians, andcomputational biologists—to address
themyr-iad questions and unmet medical needs facedby
clinicians.
It is difficult to establish a platform for thereal-time,
respectful, and effective exchange ofideas among specialists who
are usually sep-arated by time, space, methods, attitudes,
andterminology (language). The hackathon pro-
www.ScienceTranslationalMedicin
vides just such a platform for initial conceptionanddesignof a
study aswell as subsequent anal-ysis and publication (in the
literal sense ofmaking the results “public”). The hackathonalso
provides a real-time infrastructure forthe kind of simultaneous
translationof specialtyterminologies and jargon that is necessary
foreffective communication and cooperation. Last,follow-up
hackathons—events that pick up andadvance prior uncompleted
projects—offer aform of continuous peer review thatmight pro-duce
more reliable research either through rep-lication of published
results or by virtue of theirdiffering perspectives on study
conception anddesign.
HACKATHON FOR DATA ANALYSISOriginally the brainchild of
SiliconValley, CA,hackathons have proven to be successful mod-els
for innovation in business settings and aretypically organized as
intense, short-duration,competitions in which teams generate
innova-tive solutions (5). The hackathon model inte-grates
collaboration, idea generation, and grouplearning by joining
various stakeholders in amutually supportive setting for a limited
peri-od of time.
For health data analysis, the goal of thehackathon is to
assemble clinical experts, datascientists, statisticians, and
thosewith domain-specific knowledge to create ideas and
produceclinically relevant research that reduces oreliminates
biases, relies on sound statistical rig-or and adequate data
samples, and aims toproduce replicable results. For that purpose,we
have coined the term “datathon” as a port-manteau of data +
hackathon, accentuatingthe application of the hackathonmodel to
dataanalytics. For example, a critical-care datathonis an event in
which participants are broughttogether to form interdisciplinary
teams andanswer research questions in the field of criti-cal (that
is, intensive) care. In addition, usingthe term datathon avoids the
potentially neg-ative connotation of the term hackathon in theminds
of some participants and observers.
Although this approachmight bemore dif-ficult for some study
designs, the analysis ofhealth data is especially suited to benefit
fromthis model, because the data have already beencollected and are
readily available to research-ers in amachine-readable
format.Whereas theprocess flow in translational medicine is
con-ventionally from the research bench to thebedside, the flow in
the datathon model isfrom the point of care to research, ormore
spe-cifically, to the database and the analyticalteam. Approaching
health care in this manner
e.org 6 April 2016 Vol 8 Issue 333 333ps8 1
http://stm.sciencemag.org/
-
P ER SP EC T I V E
by guest on Novem
ber 23, 2017http://stm
.sciencemag.org/
Dow
nloaded from
is a form of reverse engineering, because itexamines the data
record of the system inputsand elements that produce the observed
out-comes. As databases become larger and moredetailed, they will
come to represent more andmore accurate depictions of a kind of
virtualclinical reality providing even richer resourcesfor
research.
Health care data admittedly are subject tointrinsic and
extrinsic problems of accuracy.Therefore, analytical studies based
on thesecondary use of health care data inherentlysuffer this
particular fragility. The data canbe wrong for a large variety of
reasons, includ-ing human and machine misentry, missingdata, and
faulty assessments, among others.However, all scientific research
suffers fromsome degree of data unreliability—those whohave worked
in wet labs understand that notall reported results reflect
repeatable experi-mental perfection. One further advantage ofa team
approach is that faulty data potentiallycan be recognized from a
variety of viewpointaxes by the various expert types required to
doso. For example, there may be technical prob-lems whose
recognition requires a data scientistor software engineer, clinical
issues requiringa clinician, and other issues best recognized
bydomain experts.
AN EXEMPLAR DATABASEThe MIMIC (Medical Information Mart
forIntensive Care) database contains patient-leveldata for
intensive care unit (ICU) admissionsat Beth Israel Deaconess
Medical Center from2001 to 2012 (6). Developed and maintainedby the
Laboratory of Computational Physiolo-gy at MIT, the publically
accessible databasecontains anonymized data from more than60,000
ICU admissions, with data storedacross multiple tables and
thousands of fields.
In order to give all teams in the datathonsimultaneous and
continuous access to theMIMIC database, the data were stored in
theSAP HANA in-memory relational databasemanagement system and
hosted on cloudservers. Database instances were deployeddynamically
according to the data querydemands of participants to allow for
moreefficient queries.
Note that the current MIMIC databaseserves as a model for the
clinical databases ofthe future, which will grow more completeand,
therefore, more useful as electronic healthrecords become nearly
universally imple-mented and an open, shared data philosophybecomes
more widespread. Already, discus-sions are ongoing to build an
international
consortium that leverages data from ICUs intheUnited States,
theUnitedKingdom, France,Belgium, Brazil, Japan, Australia, and
NewZealand. Other specialties are following suit:in April 2015, the
American Heart Associationconvened a 1.5-day forum to discuss
criticalissues in the acquisition, analysis, and sharingof data in
the field of cardiovascular and strokescience (7).
CRITICAL CARE DATATHONOur group hosted International Critical
Caredatathons in September 2014 and September2015. The 2014
datathon spanned three daysand three countries, with teams located
inCambridge,MA; London; andParis.More than200 participants across
all locations gathered toconduct secondary data analysis using
theMIMIC database. The organizing committeemade a particular effort
to prepare participantsin advance of the event. Video lectures and
on-line tutorials were available on the eventwebsiteto educate
participants about the MIMICdatabase, including guidelines to
define clini-cal questions and tips to extract variables.
The introductory Friday evening sessionbegan with an overview of
expectations fortheweekend; a list of dos anddon’tswere
high-lighted to make the weekend as productive aspossible.
Participants were encouraged to col-laborate, fail fast, and
iterate. This session wasfollowed by problem-based pitches, in
whichparticipants shared clinical problems andknowledge gaps that
could be addressed usingthe MIMIC database. After problem
pitching,participants divided into specifically diverseteams during
a stand-up dinner. Teams wererequired to have at a minimum one
clinician,one data engineer, and one data scientist. Wealso
encouraged participants to promote theevent on Twitter (with labels
#criticaldataand #criticaldata2014). The session endedwithan
overview of the MIMIC database and ahands-on tutorial, which helped
minimizethe technical hurdles related to software instal-lation,
cloud server connection, and program-ming syntax.
All of Saturday and part of Sunday werespent “hacking.” During
the event, partici-pants were encouraged to share code (for
ex-ample, SQL, Python, R, SAS, and STATA)used for data extraction
and statistical analysisin a GitHub repository. Use of the GitHub
re-pository served multiple purposes: (i) facili-tated code review
among members of thesame team; (ii) facilitated interteam code
re-view to cross-check methods for similarlythemed projects (8, 9);
and (iii) allowed code
www.ScienceTranslationalMedicin
to be reused for future related studies, whilepermitting
customizations to be made basedonmodifications in a study design.
On Sundayafternoon, preliminary findings were presentedby each team
and focused on the difficultiesencountered and insights gained
during thecourse of the weekend. However, the threesimultaneous
events unfolded with subtle dif-ferences and have provided insight
for im-provement in future datathons.
The 2014 Cambridge event saw more than100 registrantswith awide
rangeof self-identifiedprofessions, including
biomedical/computerengineers (n = 37), physicians (n = 33),
datascientists (n = 33), entrepreneurs (n = 22), bio-statisticians
(n = 11), patients (n = 4), nurses(n = 3), and pharmacists (n = 2).
The presenceof four patients was singular to the Cambridgehackathon
and demonstrated a valuable op-portunity for researchers to include
a popula-tion usually relegated to study participation assubjects.
In addition, the Cambridge event in-cluded some experienced
participants whohad previously attended other hackathons,
in-cluding our inaugural, single-site MIT CriticalCare datathon in
January 2014, fromwhichwegained valuable experience in the
planningand execution of an event of this kind.
In London, more than 40 participantsformed seven
interdisciplinary teams. Researchtopics included examination of the
relationshipbetween lactic acidosis andmortality, the asso-ciation
between supranormal oxygen levels inthe blood and survival after
subarachnoidhemorrhage, and visualization of outcomesamong obese
patients in the ICU. The eventalso had a number of invited speakers
who dis-cussed opportunities arising from the in-creasing
availability of data in health care.
The Paris event was structured to ensurethat the relationships
formed at the datathonwould transition into long-term
collabora-tions. To accomplish this, five teams werelimited to
three participants each, and each teamincluded one intensivist, one
biostatistician, andone data engineer. Instead of
problem-pitching,the clinical questions were selected
frompreviously submitted ideas. Last, in the spiritof continuous
collaboration and peer re-view, the Paris-based organizing
committeescheduled regular follow-up meetings andprogress
presentations with the participat-ing teams.
TheMIT Critical Care datathon has servedas amodel for and
ushered in similar events inother specialties. The following year,
in June2015, the Cross Neurodegenerative Diseasesdatathon was held
in Boston, MA (10), and
e.org 6 April 2016 Vol 8 Issue 333 333ps8 2
http://stm.sciencemag.org/
-
P ER SP EC T I V E
by guest on Novem
ber 23, 2017http://stm
.sciencemag.org/
Dow
nloaded from
focused on developing approaches to under-stand similarities and
differences across a vari-ety of neurodegenerative diseases. About
30scientists from leading institutions aroundthe world participated
in five teams, whichhad access to public datasets of interest to
re-search teams studying Parkinson’s and otherneuromuscular
diseases.
In September 2015, the follow-up secondinternational MIT
Critical Care datathon washeld, bringing back participants from the
pre-vious year’s event as well as new ones. Asoftware platform
(Fig. 1) was provided thatintegrated with Jupyter Notebook and
includ-ed a custom Python package to facilitate dataprocessing and
analysis (11). The platformconnected directly to the database,
alloweddocumentation and information sharingamong the team members
as regards the re-search question and study design, and,
mostimportantly, facilitated archiving and sharingof queries and
codes among the teams for co-hort selection, variable extraction,
and data vi-sualization and analysis. For example, teamsassisted
each other in identifying vasopressorand mechanical ventilation
use, their dosesor settings, and durations of use. Last, we
en-couraged the teams to publish their project no-tebooks at the
MIMIC website once theypublish their manuscripts, in order to
sharetheir patient cohorts, queries, and codes usedfor
analysis.
Whereas the objective of the first Interna-tional Critical Care
datathon was to draw datascientists and frontline clinicians to a
researchcommunity aroundMIMIC, the second inter-national event
focused on attracting partici-pants who are committed to publishing
theirfindings. The first datathon produced onepublication (12), and
another is currently un-der review. The second datathon is
nowpoisedto deliver its goal: Six out of the 10 teamssubmitted
abstracts of their projects for pre-sentation at the 2016 American
Thoracic Soci-ety meeting in San Francisco, CA. All sixprojects
were accepted.
LESSONS LEARNEDBiomedical research is often undertaken by
in-vestigators operating independently andworking in isolation
(13). For experimentssuch as double-blind randomized
controlledtrials, independence and isolation are criticalmeasures
to reduce bias. However, theoveremphasis on isolation and
independencein all types of research has come at the costof
validity and reproducibility. This problemis only exacerbated by
the academic reward
A
B
C
Fig. 1. Sample snapshots. Shown are screen shots of the Jupyter
Notebooks used in the SecondInternational Critical Care datathon to
facilitate sharing of queries and codes among the teams. (A)
Acustom Python package was developed to assist with transforming
and analyzing data. (B) Notebookssimplified data exploration. (C)
Data visualization.
www.ScienceTranslationalMedicine.org 6 April 2016 Vol 8 Issue
333 333ps8 3
http://stm.sciencemag.org/
-
P ER SP EC T I V E
by guest on Novem
ber 23, 2017http://stm
.sciencemag.org/
Dow
nloaded from
system, which is largely tailored toward thepromotion of
individual researchers ormultiple researchers working in the same
areaof basic science, rather than work by inter-disciplinary teams.
Further, the current aca-demic system effectively hinders
potentiallybeneficial idea generation and collaborationacross
disparate entities, because data sharingrestraints keep valuable
data within hostinstitutions. In contrast, our datathons aimto
accelerate the discovery of evidence-basedknowledge, increase
collaboration via an open,cross teams−based approach during the
designphase of research, and create a system ofiterative peer
review that might improve thereliability of research.
The importance of addressing communi-cation failure and a lack
of shared responsibil-ity among collaborators was evident in
anevent involving a Duke laboratory that joltedthe researchworld
(14). Flawed gene-expressiontests developed in the laboratory were
used inthree clinical trials to determinewhich chemo-therapy
treatment patients with lung or breastcancer would receive.
Throughout this pro-cess, the responsibilities of the
co-investigatorson the research team and lines of accountabil-ity
were apparently unclear. The U.S. Instituteof Medicine, in
response, published a manu-script in 2012 entitled Evolution of
TranslationalOmics: Lessons Learned and the Path Forward(15). The
report emphasized that given thecomplexity and multidisciplinary
nature ofomics research, there should be heightenedattention in
promoting a culture of teamworkacross disciplines, in addition to
scientific in-tegrity and transparency.
Last, a potential advantage of upfront andongoing collaborative
interdisciplinary reviewis better recognition of which ideas are
clearlyworth investigation, which are interesting butnot
particularly promising, and which are ex-tremely unlikely to be
successfully pursued. Attimes, the latter are just the kinds of
ideas thatare new and innovative and should thereforebe pursued.
But at other times, wild ideasmight simply lead to dead ends and
wasted ef-forts. Team collaboration and discussion, inaddition to
the group’s development of ideasin the first place, should help in
determiningwhether such ideas are worthy of the effortinvolved to
investigate them. With the po-tential for more efficiency in
research, theremight also be the opportunity for pilot efforts,in
which unlikely but interesting ideas are in-vestigated at a smaller
scale by a select smallerteam broken off from the main group for
thispurpose. Overall, the goal is to pursue impor-
tant and interesting ideas that seem likely topay off but also
not to neglect unusual ideasthat might ultimately be correct.
DOING DATATHONS RIGHTWith the increasing complexity of research
ina world where the low-hanging fruit has beenpicked and the issues
are often at the scale of“big data,” research has already to some
extentbecome a team-based process. We proposethat multi-expertise
viewpoints inserted open-ly into the processwould aid in the
conception,development, processing, and publication ofresearch that
is more reliable while hopefullyremaining as interesting,
innovative, and im-portant as that produced by the current sys-tem.
The datathon approach offers potentialsolutions to research
problems such as poorupfront study design with inadequate
statisticsand an insufficient number of data samples. Itmight
alsoprovide thebenefits ofmore andcon-tinuous transparency, as the
input of many andmore objective “eyes” is applied to the
entireenterprise; this could lead to more objectiveand valid
analyses and contents being providedto those in the next stage of
peer review.
Although our datathons have not yet in-volved basic biomedical
scientists, participa-tion in these kinds of events might
providethis group with ideas and opportunities. Theearly datathons
have already given data-scienceexperts a critical toehold in the
space andallowed them to leverage their skills whileexploring their
evolving interests in the area.The datathons have also given a
variety of cli-nicians normally excluded from the researchprocess
the opportunity to lend their practicalexperiential insights and
have opened up theresearch arena to entirely new and valuablegroups
such as entrepreneurs and patients. Fu-ture datathons are likely to
build and expand onthese early innovations in collaborative
research.
If half of reported research is indeed un-reliable (3), those
who read the literature arenot only wasting half their time, they
are actu-ally poisoning half their time with falsehoods.There is an
unacceptable time delay in thefeedback loop between reported
researchresults and the subsequent discovery that theresults are
not valid. In fact, the results mightnever even be carefully
scrutinized and remainas unknown falsehoods in
academic-literaturelimbo. It is important to deliver a
researchoutput that clinicians and patients can relyon to the
extent that everything we read (in-cluding this paper) should be
fundamentallyreliable, while still approached with
criticalskepticism.
www.ScienceTranslationalMedicin
We are not so naïve as to believe that thiskind of paradigm
change will be easy or un-challenged. But the current level of
researchunreliability is unacceptable and is only likelyto increase
with the deluge of digital healthdata unless the fundamental
fragilities of theresearch system are addressed.We also under-stand
that, while the proposed datathonmodelof collaborative work might
yield better sci-ence andbetter use of interdisciplinary person-nel
and resources, it is unlikely to have realimpact until the academic
reward systemchanges. The current research enterprise isan open
control loop in which the output isnot continuously evaluated for
its impact onthe plant. Academic pressures, sometimescompoundedby
influential funding bynondis-interested commercial parties, tend to
movethe research process relentlessly forward witha speed that at
times compromises due dili-gence. A datathon model with its
interdis-ciplinary team approach has the potential toraise and
answer important new questionswhile potentially reducing the
wasteful un-reliability of the current system.
REFERENCES AND NOTES1. F. Godlee, Research misconduct is
widespread and harms
patients. BMJ 344, e14 (2012).
2. A.-W. Chan, F. Song, A. Vickers, T. Jefferson, K.
Dickersin,P. C. Gøtzsche, H. M. Krumholz, D. Ghersi, H. B. van
derWorp,Increasing value and reducing waste: Addressing
in-accessible research. Lancet 383, 257–266 (2014).
3. J. P. Ioannidis, How to make more published researchtrue.
PLOS Med. 11, e1001747 (2014).
4. M. R. Macleod, S. Michie, I. Roberts, U. Dirnagl, I.
Chalmers,J. P. Ioannidis, R. Al-Shahi Salman, A.-W. Chan, P.
Glasziou,Biomedical research: Increasing value, reducing
waste.Lancet 383, 101–104 (2014).
5. E. T. Moseley, D. J. Hsu, D. J. Stone, L. A. Celi, Beyond
openbig data: Addressing unreliable research. J. Med. InternetRes.
16, e259 (2014).
6. L. A. Celi, R. G. Mark, D. J. Stone, R. A. Montgomery,
“Bigdata” in the intensive care unit. Closing the data loop.Am. J.
Respir. Crit. Care Med. 187, 1157–1160 (2013).
7. E. M. Antman, E. J. Benjamin, R. A. Harrington, S. R.
Houser,E. D. Peterson, M. A. Bauman, N. Brown, V. Bufalino, R. M.
Califf,M. A. Creager, A. Daugherty, D. L. Demets, B. P. Dennis,S.
Ebadollahi, M. Jessup, M. S. Lauer, B. Lo, C. A. MacRae,M. V.
McConnell, A. T. McCray, M. M. Mello, E. Mueller,J. W. Newburger,
S. Okun, M. Packer, A. Philippakis, P. Ping,P. Prasoon, V. L.
Roger, S. Singer, R. Temple, M. B. Turner,K. Vigilante, J. Warner,
P. Wayte; American Heart Associ-ation Data Sharing Summit
Attendees, Acquisition, anal-ysis, and sharing of data in 2015 and
beyond: A survey ofthe landscape: A conference report from the
AmericanHeart Association data summit 2015. J. Am. Heart Assoc.4,
e002810 (2015).
8. D. C. Ince, L. Hatton, J. Graham-Cumming, The case foropen
computer programs. Nature 482, 485–488 (2012).
9. G. Wilson, D. A. Aruliah, C. T. Brown, N. P. Chue Hong,M.
Davis, R. T. Guy, S. H. Haddock, K. D. Huff, I. M. Mitchell,M. D.
Plumbley, B. Waugh, E. P. White, P. Wilson, Best
e.org 6 April 2016 Vol 8 Issue 333 333ps8 4
http://stm.sciencemag.org/
-
P ER SP EC T I V E
practices for scientific computing. PLOS Biol. 12,
e1001745(2014).
10. C. I. Barash, K. Elliston, R. Potenzone, tranSMART
Founda-tion Datathon 1.0: The cross neurodegenerative
diseaseschallenge. Applied Transl. Genomics 6, 42–44 (2015).
11. F. Pérez, B. E. Granger, IPython: A system for
interactivescientific computing. Comput. Sci. Eng. 9, 21–29
(2007).
12. J. D. Salciccioli, D. C. Marshall, M. A. Pimentel, M. D.
Santos,T. Pollard, L. A. Celi, J. Shalhoub, The association
betweenthe neutrophil-to-lymphocyte ratio and mortality in
crit-ical illness: An observational cohort study. Crit. Care 19,
13(2015).
13. J. P. Ioannidis, S. Greenland, M. A. Hlatky, M. J. Khoury,M.
R. Macleod, D. Moher, K. F. Schulz, R. Tibshirani,Increasing value
and reducing waste in research design,conduct, and analysis. Lancet
383, 166–175 (2014).
14. The Economist (2011) An array of errors. 10th September
2011.15. C. M. Micheel, S. J. Nass, G. S. Omenn, Evolution of
Translational Omics: Lessons Learned and the Path For-ward
(National Academies Press, Washington, DC, 2012).
Acknowledgments: The MIMIC database is funded by the U.S.
National Institutes of Health grant R01 EB001659. The MITCritical
Care datathons were funded by the MIT International
www.ScienceTranslationalMedicin
Science and Technology Initiatives. The London datathonwas
supported by The IET: The Institution of Engineeringand Technology.
Competing interests: The authors declarethat they have no competing
interests.
10.1126/scitranslmed.aad9072
Citation: J. Aboab, L. A. Celi , P. Charlton, M. Feng,M.
Ghassemi, D. C. Marshall, L. Mayaud, T. Naumann,N. McCague, K. E.
Paik, T. J. Pollard, M. Resche-Rigon,J. D. Salciccioli, D. J.
Stone, A “datathon” model to supportcross-disciplinary
collaboration. Sci. Transl. Med. 8, 333ps8 (2016).
e.org 6 April 2016 Vol 8 Issue 333 333ps8 5
by guest on Novem
ber 23, 2017http://stm
.sciencemag.org/
Dow
nloaded from
http://stm.sciencemag.org/
-
A ''datathon'' model to support cross-disciplinary
collaboration
Salciccioli and David J. StoneMayaud, Tristan Naumann, Ned
McCague, Kenneth E. Paik, Tom J. Pollard, Matthieu Resche-Rigon,
Justin D. Jerôme Aboab, Leo Anthony Celi, Peter Charlton, Mengling
Feng, Mohammad Ghassemi, Dominic C. Marshall, Louis
DOI: 10.1126/scitranslmed.aad9072, 333ps8333ps8.8Sci Transl
Med
ARTICLE TOOLS http://stm.sciencemag.org/content/8/333/333ps8
CONTENTRELATED
http://stm.sciencemag.org/content/scitransmed/8/338/338ed6.fullhttp://stm.sciencemag.org/content/scitransmed/8/336/336ed5.fullhttp://stm.sciencemag.org/content/scitransmed/8/335/335ps10.full
REFERENCES
http://stm.sciencemag.org/content/8/333/333ps8#BIBLThis article
cites 13 articles, 2 of which you can access for free
PERMISSIONS
http://www.sciencemag.org/help/reprints-and-permissions
Terms of ServiceUse of this article is subject to the
is a registered trademark of AAAS.Science Translational
Medicinetitle licensee American Association for the Advancement of
Science. No claim to original U.S. Government Works. TheScience,
1200 New York Avenue NW, Washington, DC 20005. 2017 © The Authors,
some rights reserved; exclusive
(ISSN 1946-6242) is published by the American Association for
the Advancement ofScience Translational Medicine
by guest on Novem
ber 23, 2017http://stm
.sciencemag.org/
Dow
nloaded from
http://stm.sciencemag.org/content/8/333/333ps8http://stm.sciencemag.org/content/scitransmed/8/335/335ps10.fullhttp://stm.sciencemag.org/content/scitransmed/8/336/336ed5.fullhttp://stm.sciencemag.org/content/scitransmed/8/338/338ed6.fullhttp://stm.sciencemag.org/content/8/333/333ps8#BIBLhttp://www.sciencemag.org/help/reprints-and-permissionshttp://www.sciencemag.org/about/terms-servicehttp://stm.sciencemag.org/
