Websites

1
Websites

• http//mor.nlm.nih.gov/download/rxnav/
• http//www.stccmop.org/quarry

2
Profiling DataspacesUnderstanding (and Using)
Other Peoples Data

• David Maier
• Department of Computer Science
• Portland State University

3
With Much Support

• Dataspaces Alon Halevy, Mike Franklin
• RxSafe Paul Gorman, Karl Ordelheide, Judy Logan,
  Nick Rayner (InfoSonde)
• SACO Shannon McWeeney, Ranjani Ramakrishnan
• Quarry Bill Howe, James Rucker
• DIESEL Lois Delcambre, David Archer, Susan
  Price, Scott Fletcher, John McCall
• Funding NSF, AHRQ, DARPA

4
Other Peoples Data

• Why is other peoples data hard to understand?
• They arent around to explain it
• What they wrote down about it wasnt quite true
  or complete in the first place
• The kinds of data in a source have expanded
  beyond the original intent schema drift
• How do you understand the data?

5
Dataspaces

• Deal with all the data from an enterprise in
  whatever models
• Data co-existence
• Might not be fully integrated, especially early
  on
• Pay-as-you-go services
• Im interested in understanding sources and their
  relationships

From Databases to Dataspaces A New
Abstraction for Information Management, Michael
Franklin, Alon Halevy, David Maier, SIGMOD
Record, December 2005.
6
Example Dataspace RxSafe

• Consolidated medication list for rural elders
• Points in lifetime of a prescription
• Order (clinic, hospital)
• Dispensing (pharmacy)
• Approval (insurer)
• Administration (rehabilitation facility)
• Relevant Standards
• NDCD, RxNorm, NDF-RT

7
NDCD National Drug Code Directory

• Codes for drug packages

8
Sample NDC 62584-023-00
9
RxNorm Drug Nomenclature

• RxNav from National Library of Medicine

10
NDF-RT

• National Drug File Reference Terminology
• From Veterans Affairs
• Drug class
• Chemical class
• Effects and actions

11
NDF-RT (Blue)
12
Goals of Understanding Based on Intended Purpose

• Grouping similar medications
• Connecting possible incarnations of same
  prescription
• Generic Brand Name
• Combining medication information for a given
  patient
• Must be error preserving

13
What You Know is Wrong

• NDC and RxNorm talking about same things
• NDC tradenames 18913
• RxNorm brand names 7600
• Strings in common 418
• All RxNorm relationships have explicit inverses

14
What You Know is Incomplete
15
Incomplete Description
Source National Library of Medicine
Doesnt mention atoms, attributes Doesnt include
SY, ET, OCD, OBD
16
You Might Misunderstand What Youre Told

• RxNorm diagram is for instances
• Multi-ingredient drug case not covered

17
The Structure Isnt What You Thought

• RxNorm uses UMLS, not domain-specific
• More complex than this can have several atoms
  in each concept

18
How Do You Make Progress?

• Tools with minimal assumptions
• Ability to check hypothesis
• Means to customize data to intended task

19
What I WantDataspace Charting Toolkit

• Help with Familiarization, Profiling, Enhancement
• Inspector for generic models
• Dataspace profiler
• Assumption tracker and checker
• Structure discovery techniques
• Customization to task based on discovered
  characteristics

20
Quarry Data Model (Howe)

• resource, property, value
• (subject, predicate, object) if you prefer
• no intrinsic distinction between literal values
  and resource values
• no explicit types or classes

21
Example RxNorm
Concept
Relationship
Atom
userkey
prop
value
10001
NDC
1
10001
ORIG_CODE
123
10001
ingredient_of
10004
10001
type
DC
up to 23M triples describing 0.6M concepts and
atoms
22
Example Metadata for Scientific Data Repository
/anim-sal_estuary_7.gif
23
SKIP
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
Quarry API

• /2004/2004-001//anim-tem_estuary_bottom.gif
• aggregate bottom
• animation isotem
• day 001
• directory images
• plottype isotem
• region estuary
• runid 2004-001
• year 2004
• /2004/2004-001//amp_plume_2d.gif
• day 001
• directory images
• plottype 2d
• region plume
• runid 2004-001
• year 2004

32
Behind the Scenes

• Signatures
• resources possessing the same properties
  clustered together
• Posit that Signatures ltlt Resources
• Queries evaluated over Signature Extents

33
Experimental Results

• Yet Another RDF Store
• Several B-Tree indexes to support
• spo, po ? s, os ? p, etc.
• 3M triples
• We looked at multi-term queries

?s ltp0gt lto0gt ?s ltp1gt lto1gt ?s ltpngt ltongt
34
Experimental Results Queries
3.6M triples 606k resources 149 signatures
35
Dataspace ProfilingNDC Examples (Rayner)

• l_seq_no is key of listings yes
• lblcode, prodcode key of listings no
• 45,953 45,972
  (19)
• firm_name ? lblcode no
• 2931 2952 (21)
• each product listing should have gt0 packages and
  gt0 ingredients
• 44,972 (1180) 45,180 (792)

36
Checking Across SourcesNDC vs. RxNorm

• Ingredients
• 2794 ingredient names in NDCD
• 5145 ingredients in RxNorm
• 1570 equal strings

37
What to Do with Flawed Assumptions?

• Track exceptions
• Refine assumption
• firm_name, location ? lblcode
• Refine knowledge of world
• RxNorm has ingredient variants (which have the
  same type as ingredients)
• Want to track assumptions as they evolve, results
  of checks

38
Customization InfoSonde

• Support customizations appropriate to discovered
  data characteristics
• Three-part modules
• Probe Check or discover properties
• Switch Present applicable customizations
• Check Test that chosen switch is still valid

39
Linkage Extension Module
Want to extend a linkage based on a discovered
functional relationship Probe Join satisfies
FD here, ING ? DC Switch Materialize functional
relationship can be used to extend original
relation with DC Check Test that FD still holds
40
My First Visit to Zürich

• See
• http//www.cs.pdx.edu/maier/france-report/report1
  3.txt