This presentation describes several deficiencies of drug product labels and how information on the semantic web can be used to update the drug product label.
Circulatory Shock, types and stages, compensatory mechanisms
Boyce cshals-2012-linked-sp ls-final
1. Dynamic Enhancement
of Drug Product Labels
Through Semantic Web
Technologies
(A W3C HCLS IG Use Case)
Richard Boyce, University of Pittsburgh
Maria Liakata, Aberystwyh University/EBI
Jodi Schneider, DERI, Galway
Anita de Waard, Elsevier
1 Biomedical Informatics
Department of Biomedical Informatics
2. Overview
• The motivation for the W3C Use Case
• What we are suggesting
• How it can be accomplished
– Proof of concept
• Issues to be addressed
• Discussion
2 Biomedical Informatics
3. The problem
• Much of the information in drug package
inserts (aka product labels) is incomplete
or has not been updated
3 Biomedical Informatics
4. Missing drug-drug interactions
• Example:
– “Both clopidogrel and ticlopidine significantly inhibited the
CYP2B6-catalyzed bupropion hydroxylation. Patients
receiving either clopidogrel or ticlopidine are likely to require
dose adjustments when treated with drugs primarily
metabolized by CYP2B6.” [1]
• Search bupropion drug package inserts for
“clopidogrel”
– No mention at all in Aplenzin ER insert [2]
– Mention in the generic tablet insert [3], but refers only
to hypothetical interaction
1. Turpeinen M, Tolonen A, Uusitalo J, Jalonen J, Pelkonen O, Laine K. Effect of clopidogrel and ticlopidine on cytochrome P450 2B6 activity as measured by
bupropion hydroxylation. Clin Pharmacol Ther. 2005 Jun;77(6):553-9.
2. http://dailymed.nlm.nih.gov/dailymed/lookup.cfm?setid=2315d6e5-144d-4163-9711-f855c759cf8e
3. http://dailymed.nlm.nih.gov/dailymed/lookup.cfm?setid=e537ed50-6677-45eb-9e62-af62d3831f19
4 Biomedical Informatics
5. Missing metabolic properties
• Cimetidine inhibition of CYP3A4 and CYP1A2
metabolism
– Noted in an FDA draft guidance [1] but not PI for the
generic tablet [2]
• Fluconazole inhibition of CYP3A4 metabolism
– Noted in an FDA draft guidance [1] but not directly in
the fluconazole solution PI [3]
• Inferable from the stated interaction with midazolam
• See others like this in [4]
1. FDA. FDA Guideline: Drug Interaction Studies – Study Design, Data Analysis, and Implications for Dosing and Labeling. Rockville, MD: Food and Drug
Administration; 2006. Available at: http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/ucm072101.pdf.
2. http://dailymed.nlm.nih.gov/dailymed/lookup.cfm?setid=3a2f773d-b67f-4e83-b8d7-fce6b2887efe
3. http://dailymed.nlm.nih.gov/dailymed/lookup.cfm?setid=26672a97-adf3-9f1d-4bdc-b2606747bbf5
4. Boyce R, Harkema H, Conway M. Leveraging the semantic web and natural language processing to enhance drug-mechanism knowledge in drug product labels.
In: Proceedings of the 1st ACM International Health Informatics Symposium. IHI ’10. New York, NY, USA: ACM; 2010:492–496. Available at:
http://doi.acm.org/10.1145/1882992.1883070.
5 Biomedical Informatics
6. Missing age-related clearance data
• The PI for citalopram [1] notes age-related
pharmacokinetic changes
– “…subjects ≥ 60 years of age were compared to younger
subjects in two normal volunteer studies. In a single-dose study,
citalopram AUC and half-life were increased in the elderly
subjects by 30% and 50%, respectively”
• However, clearance (Cl) would be the preferred measure
of age-related change – is there literature on this?
– Yes! [2-4]
• The package inserts rarely have this information even
when published [5]
1. http://dailymed.nlm.nih.gov/dailymed/lookup.cfm?setid=4259d9b1-de34-43a4-85a8-41dd214e9177
2. Reis M, Lundmark J, Bengtsson F. Therapeutic drug monitoring of racemic citalopram: a 5-year experience in Sweden, 1992-1997. Ther Drug Monit. 2003;25(2):183-
191.
3. Gutierrez M, Abramowitz W. Steady-state pharmacokinetics of citalopram in young and elderly subjects. Pharmacotherapy. 2000;20(12):1441-1447.
4. Bies RR, Feng Y, Lotrich FE, et al. Utility of sparse concentration sampling for citalopram in elderly clinical trial subjects. J Clin Pharmacol. 2004;44(12):1352-1359.
5. Boyce RD, Handler SM, Karp JF, Hanlon JT. Age-Related Changes in Antidepressant Pharmacokinetics and Potential Drug-Drug Interactions: A Comparison of
Evidence-Based Literature and Package Insert Information. Am J Geriatr Pharmacother. 2012 Jan 27. [Epub ahead of print] PubMed PMID: 22285509.
6 Biomedical Informatics
7. Is there any requirement that the inserts
be kept up to date?
• Yes - 2006 regulations explicitly require that studies of
drug-drug interactions, metabolic pathways, and special
populations be discussed [1]
1. FDA. Code of Federal Regulations 21 Part 201.56, chapter Requirements on content and format of labeling for human prescription drug and biological
products. Washington, DC: US Government Printing Office, 2010. http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/CFRSearch.cfm?fr=201.57
7 Biomedical Informatics
8. Why are package inserts important?
• Package inserts are primarily intended to be a reference
for prescribing clinicians [1]
1. Marroum PJ, Gobburu J. The product label: how pharmacokinetics and
pharmacodynamics reach the prescriber. Clin Pharmacokinet. 2002;41(3):161-169.
2. Ko, Y. et al. Prescribers’ knowledge of and sources of information for potential drug-
drug interactions: a postal survey of US prescribers. Drug Saf. 31, 525-536 (2008).
8 Biomedical Informatics
9. Why is the package insert apparently
so influential?
• Authoritative
• Simple (for a drug expert) to follow
• Often the only source of information besides
FDA approval documentation for new drugs
• No standard for searching the scientific
literature
• No standard for judging the quality of
published studies
9 Biomedical Informatics
10. What are we suggesting then?
• “Use natural language processing and scientific
discourse ontologies to build a linked open data
store of scientific information that updates or
elaborates on medication safety statements
present in drug product labels.”
- W3C HCLS IG Use Case (http://goo.gl/wP4Mz)
10 Biomedical Informatics
16. How can this be accomplished?
16 Biomedical Informatics
17. The Structured Product Label (SPL)
standard makes this possible
• All package inserts for currently marketed drugs
are available in this format [1-3]
1. http://www.fda.gov/OHRMS/DOCKETS/98fr/FDA-2005-N-0464-gdl.pdf
2. http://www.fda.gov/ForIndustry/DataStandards/StructuredProductLabeling/default.htm
3. http://dailymed.nlm.nih.gov/dailymed/downloadLabels.cfm
17 Biomedical Informatics
18. FDA law dictates the kinds of claims
that should be present in each section
18 Biomedical Informatics
19. Many of these claims can be identified in
resources on the Semantic Web….
Drug Interaction Knowledge Base : http://thedatahub.org/dataset/linked-structured-product-labels
ClinicalTrials.gov: http://thedatahub.org/dataset/linkedct
DrugBank: http://thedatahub.org/dataset/the-drug-interaction-knowledge-base
19 Biomedical Informatics
20. …and then linked to package insert
sections published on the Semantic Web
Linked SPLs: http://thedatahub.org/dataset/linked-structured-product-labels
20 Biomedical Informatics
22. Overview of the proof of concept
• http://goo.gl/rs3Fy
– Package inserts for drug products containing
citalopram and venlafaxine
– Created using three Semantic Web nodes
• http://thedatahub.org/dataset/linked-structured-product-labels
• http://thedatahub.org/dataset/linkedct
• http://thedatahub.org/dataset/the-drug-interaction-knowledge-base
22 Biomedical Informatics
23. More about the data nodes
• http://goo.gl/rs3Fy
1. http://thedatahub.org/dataset/linked-structured-product-labels
2. http://thedatahub.org/dataset/linkedct
3. http://thedatahub.org/dataset/the-drug-interaction-knowledge-base
23 Biomedical Informatics
25. Take the drug interactions section of a
drug package insert…
25 Biomedical Informatics
26. Make it simple for the package insert
reader to see claims that could expand or
update the information in this section…
26 Biomedical Informatics
27. Example: an interaction affecting venlafaxine that may
not be in this section…
27 Biomedical Informatics
28. Example: an interaction affecting venlafaxine that may
not be in this section…
NLP determined that evidence for a
diphenhydramine/venlafaxine interaction is not
mentioned
28 Biomedical Informatics
29. Clicking on the link provides a path to
evidence for (or against) the drug interaction
claim…
30. One goal is to link directly to statements
and sections within scientific documents
from which evidence was derived
• For example, assume a drug interaction claim
– identify core scientific concepts (CoreSC) in the
paper containing the evidence for/against the claim
(background, methods, results,…)
– Identify the text containing evidence for/against a
claim within the CoreSC
– Link the specific CoreSC element back to the
package insert section (via the interaction claim)
• Currently testing this approach using SAPIENTA
(www.sapientaproject.com)
31. Just a couple of the issues to be
addressed
• Ensuring the quality of the claims linked to the
package insert sections
– Provenance is key
– Would nanopublications help here?
• The sky could be the limit if a drug information “knowledge
market” could be created
• Technical issues with some existing linked
open drug data nodes
– Now is the time to improve linked open drug data
• Correct RDF mappings, accurate encodings
• Graph and data provenance
31 Biomedical Informatics
32. Want more information?
• Use Case description
– http://goo.gl/wP4Mz
• Google code project
– code.google.com/p/swat-4-med-safety/
• Proof of concept
– http://goo.gl/rs3Fy
• Linked data nodes used in the proof of concept
– http://thedatahub.org/dataset/linked-structured-product-labels
– http://thedatahub.org/dataset/the-drug-interaction-knowledge-base
– http://thedatahub.org/dataset/linkedct
32 Biomedical Informatics
33. Acknowledgements
• The Drug Interaction Knowledge Base team
– John Horn Pharm.D, Carol Collins MD, Greg
Gardner, Rob Guzman
• W3C LODD Task Force
• Early comments on the Use Case:
– Michel Dumontier and several others who attend
W3C HCLS calls
33 Biomedical Informatics
35. The package insert’s influence is
significant
• How frequently do three drug interaction checking
tools agree on an interaction by source? (unpublished)
– 44 interactions from the package insert or scientific
literature (25 newer psychotropics)
– Agreement 3.25 times more likely for package
insert only interactions then for literature only
Scientific literature Package insert
15.4% 38.5% 50%
35 Biomedical Informatics