Towards linked vital registration data for reconstituting families and creating longitudinal health histories
- 1. Towards Linked Vital Registration Data for Reconstituting Families and Creating Longitudinal Health HistoriesLongitudinal Health Histories Oya Beyan, Ciara Breathnach, Sandra Collins, Christophe Debruyne, Stefan Decker, Dolores Grant, Rebecca Grant, and Brian Gurrin 21st of July 2014 – KR4HC Workshop – Vienna, Austria21st of July 2014 – KR4HC Workshop – Vienna, Austria
- 2. Irish Record Linkage, 1864-1913 • Developing a platform applying semantic technologies to historical birth-, death andtechnologies to historical birth-, death and marriage certificates. • Answering questions such as: “How accurate are historic maternal mortality rates (MMR) and infant mortality rates (IMR) for Dublin?” • Team consists of researchers (historians), digital archivists, and knowledge engineers. 21/07/2014 2
- 3. Data: General Office Records • Vital registration data – Birth-certificates– Birth-certificates – Death-certificates – Marriage records • Digitised TIFF images of hardcopy indexes and registers. • 2 TB of data• 2 TB of data • Database describing the digitised records allowing searches on some fields. 21/07/2014 3 ©General Records Office of Ireland 2014
- 4. Challenges • Certified causes of death that can be attributed to maternal death – Within 42 days after labour – before (1864) it was 12– Within 42 days after labour – before (1864) it was 12 – Septicemia (blood poisoning), Fever, … – “Corresponding” birth certificate? • Death certificates with no corresponding birth certificate • “Gaps” in sibship interval, even though no birth- or death certificates can be found. • The terminology used pre-1900. E.g., “debile” to denote• The terminology used pre-1900. E.g., “debile” to denote weak or a failure to thrive. • Capturing the socio-economical status of the families via, for instance, the professions, ranks of fathers. 21/07/2014 4
- 5. Conceptual Architecture Digital Archivist SPARQL endpoint / Linked Data Server Updates GRO records as RDF LinksLinker UpdaterRepository Triple- store Linked Data Server Analytics Researcher 21/07/2014 5 DATA ANALYTICSPRESERVATION Links to external datasets: e.g., Logainm – a database of Irish historical and contemporary place names to provide additional context.
- 6. Development of 2 ontologies Triplestore 2 Data Analysis CONCERNSSEPARATIONOFCONCERNS Obviously, due to the sensitive nature of the data, data protection is key. 21/07/2014 6 GRO Triplestore Transformation from one model to another • SPIN – SPARQL Inference • SWRL / RuleML • SPARQL Construct • … SEPARATION protection is key.
- 7. Development of 2 ontologies • 2 ontologies were developed – separation of concerns • First ontology for describing the contents of records – OWL 2 shallow, “flat ontology” • Second ontology for data analysis – OWL 2 + rules – Rules to capture background and domain knowledge– Rules to capture background and domain knowledge – Developed by having the historians formulate competency questions (Grüninger and Fox) – Captured graphically using Object Role Modelling 21/07/2014 7
- 8. Graphical Representation in ORM 21/07/2014 8
- 9. ### Prefixes ommitted … irl:Record a owl:Class ; rdfs:label "Record" ; . irl:Certificate a owl:Class ; rdfs:label "Certificate" ; rdfs:subClassOf irl:Record; .rdfs:subClassOf irl:Record; . irl:BirthRecord a owl:Class ; rdfs:label "Birth Record" ; rdfs:subClassOf irl:Certificate ; . irl:DeathRecord a owl:Class ; rdfs:label "Death Record" ; rdfs:subClassOf irl:Certificate ; . irl:MarriageRecord a owl:Class ; rdfs:label "Marriage Record" ;rdfs:label "Marriage Record" ; rdfs:subClassOf irl:Record ; . irl:Return a owl:Class ; rdfs:label "Return" ; . … 21/07/2014 9
- 10. Conclusions • Presented the problem and highlighted the challengeschallenges • Developed two ontologies – Encoding contents of digitized GRO records for long-term digital preservation DRI – Data analytics to answer the researchers’ question – in this case a historianquestion – in this case a historian • Data exploration and annotation of the records started on a subset of the dataset 21/07/2014 10