The FindMeEvidence project An open-source, mobile-friendly search - - PDF document

The FindMeEvidence project

An open-source, mobile-friendly search engine for public medical knowledge

Matthias SAMWALDa, Allan HANBURYb. a Medical University of Vienna, Austria b Vienna University of Technology, Austria

Many physicians access the web during patient consultations

Matthias Samwald, Marlene Kritz, Manfred Gschwandtner, Veronika Stefanov, Allan Hanbury. „An open, trustworthy and multilingual search engine for medical practitioners“ MIE2011

But they are confronted with several barriers to using current web search engines in an effective way

!

Matthias Samwald, Marlene Kritz, Manfred Gschwandtner, Veronika Stefanov, Allan Hanbury. „An open, trustworthy and multilingual search engine for medical practitioners“ MIE2011

And they would rather prefer not to pay to have these problems fixed…

!

Matthias Samwald, Marlene Kritz, Manfred Gschwandtner, Veronika Stefanov, Allan Hanbury. „An open, trustworthy and multilingual search engine for medical practitioners“ MIE2011

Insight: Improving the availability of free, open, mobile-friendly medical search engines is extremely valuable for medical professionals & quality of care!

To this end, we are developing the FindMeEvidence search engine system FindMeEvidence is an open-source medical search engine based on medical content on the web Optimized for efficient use in time-constrained medical practice Small footprint, easy to maintain and customize to local

content requirements

!

FindMeEvidence aims to be a simple, yet highly effective search engine for medical professionals

Simple user interface Enabling rapid synthesis of evidence for medical decision support

Do you also like it when the answer you searched for is already shown in the Google result snippet? It should be like that all the time.

Accessible on all devices (especially mobile) Complementing, not replacing Google & PubMed

Very simple user interface

!

Autocomplete via PubMed API

(Appears vastly more practical than all vocabulary/taxonomy-based autocomplete methodologies)

Options for result filtering are kept simple as well

Contains only English content, but offers query translation support to make content better accessible

Physicians self-reported good English skills in our Europe-wide study But active vocabulary might not be as good, so we assist query formulation

!

Where available, key findings of articles in PubMed are shown directly in search results list

!

Local abbreviations in the abstracts are expanded for improving these previews of key findings

!

In abstract (note abbreviations) In preview (note automatically expanded abbreviations; would not be legible otherwise)

Open-access signalling and linking

FindMeEvidence aims to signal the open-access status of publications to users Great aid for users without institutional access (i.e., most non-academic medical settings) Direct link to mobile-friendly articles (PMC PubReader)

Quality signalling

Wikipedia is included due to popular demand and real practical benefits (very good for getting quick

• verview)

However, quality of Wikipedia articles varies substantially Measures are taken to notify users of potential problems with content quality of certain articles

We will never replace Google and PubMed. So let‘s try to play nice with them!

Links to query results in Google / PubMed at bottom of page

Preliminary evaluation: Comparison with state-

• f-the-art search engine for evidence-based

medicine

User query Inferred information need FindMeEvidence met success criteria TRIP Database met success criteria flexible bronchoscopy General information about treatment no yes paclitaxel + doxorubicin cardiotoxicity Specific drug side effects (Were they reported? How significant are they? Can this problem be addressed?) yes yes uterine fibroids risk factors Risk factors for specific condition yes no acute liver failure General information, especially about diagnosis and treatment yes no ssri sexual General information about sexuality-related drug side effects of SSRIs yes no Viral induced wheeze General information about symptom yes yes streptococcal pharyngitis AND child Diagnosis/treatment in pediatric population yes yes mirtazapine AND hepatitis Drug side effect or information on how to tailor treatment when comorbidity present No no … … … …

Excerpt of evaluation queries. Queries were selected from search logs of TRIP Database and PubMed.

Preliminary evaluation: Search results of FindMeEvidence are competitive

! Checked if at least one result in first 5 results met information need of test query (no facets selected) Out of all test queries (N = 36), FindMeEvidence results met success criteria for 25 (69,4%) of the queries, while TRIP Database results met criteria for

• nly 17 (47,2%) of the queries.

Only a very limited preliminary evaluation! Does not proof superiority, but suggests that FindMeEvidence can be of practical use.

Wrapping up: How does all that relate to you?

!

Wrapping up: How does all that relate to you?

You can use FindMeEvidence.org search engine at your institution, or create a local installation customized to your needs! We are looking for organisations to partner in evaluating and improving the system ! "##"$%&'(%) *"$ $+,+-$./ #-#+$,0

Join open-source project on Github

!

Thanks!

Local team

• Dr. Matthias Samwald (FindMeEvidence team lead)

Georg Petz (MSc student, core developer of Version >1.0)

• Dr. Claus-Dieter Volko
• Dr. Veronika Stefanov (Khresmoi project)
• Dr. Allan Hanbury (Khresmoi project scientific coordinator)

! ! !

Web http://FindMeEvidence.org/ (currently still V1.0, will be upgraded to V1.1 in coming weeks) https://github.com/matthias-samwald/find-me-evidence (feel free to fork and/or contribute!) ! Contact matthias.samwald @ meduniwien.ac.at ! Funding European Union Seventh Framework Programme agreement No. 257528 (KHRESMOI) Austrian Science Fund (FWF): [PP 25608-N15]

!

!

Extra slides

FindMeEvidence is based on simple key technologies Apache Solr 4

Industry-strength search server with wide support and user-base

JQuery mobile

Cross-platform, mobile-friendly web user interface library

PHP

!

!

We developed a methodologies for selecting clinically relevant content from PubMed and Wikipedia

PubMed: Approx. 600 journals included, approx 700.000 articles

Selection is needed for excluding vast number of preclinical study results on PubMed

Wikipedia: Articles belonging to Wikiprojects Medicine and Pharmacology

Only highly relevant information for decision support shown in result lists

Title Journal (to judge relevance and quality) Publication date (to judge actuality) Key assertions (click needed for abstract / full text) ! Deliberately not shown:

Author names (only matters for detailed scientific research, you would use PubMed for that anyways) Keywords, tags Bibliographic ‚metacrap‘