UBC Faculty Research and Publications

Automatic detection and resolution of measurement-unit conflicts in aggregated data Samadian, Soroush; McManus, Bruce; Wilkinson, Mark May 8, 2014

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
52383-12920_2014_Article_472.pdf [ 412.22kB ]
Metadata
JSON: 52383-1.0223374.json
JSON-LD: 52383-1.0223374-ld.json
RDF/XML (Pretty): 52383-1.0223374-rdf.xml
RDF/JSON: 52383-1.0223374-rdf.json
Turtle: 52383-1.0223374-turtle.txt
N-Triples: 52383-1.0223374-rdf-ntriples.txt
Original Record: 52383-1.0223374-source.json
Full Text
52383-1.0223374-fulltext.txt
Citation
52383-1.0223374.ris

Full Text

RESEARCH Open AccessAutomatic detection and resolution ofmeasurement-unit conflicts in aggregated dataSoroush Samadian1, Bruce McManus1, Mark D Wilkinson2*From The 3rd Annual Translational Bioinformatics Conference (TBC/ISCB-Asia 2013)Seoul, Korea. 2-4 October 2013AbstractBackground: Measurement-unit conflicts are a perennial problem in integrative research domains such as clinicalmeta-analysis. As multi-national collaborations grow, as new measurement instruments appear, and as Linked OpenData infrastructures become increasingly pervasive, the number of such conflicts will similarly increase.Methods: We propose a generic approach to the problem of (a) encoding measurement units in datasets in amachine-readable manner, (b) detecting when a dataset contained mixtures of measurement units, and (c)automatically converting any conflicting units into a desired unit, as defined for a given study.Results: We utilized existing ontologies and standards for scientific data representation, measurement unitdefinition, and data manipulation to build a simple and flexible Semantic Web Service-based approach tomeasurement-unit harmonization. A cardiovascular patient cohort in which clinical measurements were recorded ina number of different units (e.g., mmHg and cmHg for blood pressure) was automatically classified into a numberof clinical phenotypes, semantically defined using different measurement units.Conclusions: We demonstrate that through a combination of semantic standards and frameworks, unit integrationproblems can be automatically detected and resolved.BackgroundIntegration, comparison and interpretation of quantita-tive data require, as a first step, that all measurementsare represented in the same units. Discordance in unitsis common in integrative research, is difficult to detect,and has severe consequences when not managed effec-tively. Even NASA has made serious and expensiveerrors by failing to detect and account for measure-ment-unit conflicts[1].This problem is well-recognized in clinical research,due to its complex, multi-dimensional and heteroge-neous nature, and where highly disparate datasets, oftenfrom non-coordinating groups, need to be broughttogether. This work, therefore, is contextualized within aclinically-oriented study in which we would be requiredto gather clinical data from a number of participatinggroups, and attempt to automatically categorize indivi-dual patients over existing health-risk guidelines usingsemantic technologies[2]. Prior to undertaking thestudy, we became aware of the potential for measure-ment unit conflicts in these integrated datasets. Ratherthan creating an ad hoc solution, we attempted to definea lightweight, standards-compliant, and semantics-basedsolution that could be re-used by other bio/medicalresearch projects.It should be noted that, in the current work, it wasnot our intention to propose novel epistemic theories ofqualities and measurements as it is both beyond thescope of our present work, and not necessarily requiredto achieve our major objectives in the current studywhich was to focus on providing practical and opera-tional solution for clinicians and other health research-ers using existing standards and theoretical frameworks.The reader more interested in detailed core theoreticalfoundations on which this study (and many existing* Correspondence: markw@illuminae.com2Centro de Biotecnología y Genómica de Plantas, Universidad Politécnica deMadrid, Madrid, EspañaFull list of author information is available at the end of the articleSamadian et al. BMC Medical Genomics 2014, 7(Suppl 1):S12http://www.biomedcentral.com/1755-8794/7/S1/S12© 2014 Samadian et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the CreativeCommons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, andreproduction in any medium, provided the original work is properly cited. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.ontologies such as DOLCE) is based, is referred to(among others) [3-5].The rest of this paper is organized as follows. We willexplain the challenges associated with formal measure-ment-unit representation and integration in the Seman-tic Web, and discuss related tools and resources. Wedescribe possible design-choices for modeling units, andthe problems or benefits of these alternatives. We thendiscuss our proposed framework, and justify the coredesign and representation choices through presentationof a case study of patient phenotype classification withinthe cardiovascular domain. Finally, concluding remarkstogether with future extensions will be presented.Related workVarious standards exist for unit representation, the mostnotable of which (in the Western world) is the Interna-tional System of Standards (SI), now adopted in all areasof science[6]. Nevertheless, the choice of units, evenwithin this system, is sufficiently broad that reliableautomated integration of quantitative measurementsremains problematic [7], and as a result, a consistentmethodology to interpret and integrate the units withinand between datasets remains to be established.Recently, semantic solutions in the form of measure-ment unit ontologies have emerged as a potential pathtowards a solution. These ontologies generally use theResource Description Framework (RDF) [8] for dataencoding and Web Ontology Language(OWL)[9] toencode axioms that can be used for automated inferenceover the data. However, while defining standards for therepresentation of numerical/quantitative data, the RDFstandard does not inherently define an approach torepresenting the measurement units associated with thatdata [10]. Additionally, these ontologies are oftendomain-specific, and therefore have limited coverage ofthe full range of measurement units - focusing only onunits relevant to that domain of investigation - and lackinformative relationships between related units. Forinstance, the GALEN concept MilligramPerDeciLiter, isdefined as a subclass of the concepts ConcentrationUnit,however it lacks any indication that this unit is com-posed of combination of two base units (gram and liter)and two prefixes (milli and deci).Recently, ontologies have been developed to moregenerically address formalizations of unit representationand integration. Prominent examples of such ontologiesare the Measurement Unit Ontology (MUO)[10], Ontol-ogy for Engineering Mathematics(EngMath)[11], Quanti-ties, Units, Dimensions and Types (QUDT)[12], and theOntology of Units of Measure (OM) [13]. Their salientfeatures are (for a more comprehensive review pleaserefer to the [14]):MUO: In MUO, complex measurement units can bederived from the base ones in a modular fashion. MUOproposes a convenient framework for defining new unitsof measurements in terms of existing ones. However,while MUO defines metric prefixes (e.g., centi- and kilo-),which could be used to automate automated conversionfor SI-based measurements, it lacks quantitative orformula-based definitions for converting between SI unitsand similar qualities in other unit-systems (e.g., inchand cm).Ontology for Engineering Mathematics(EngMath):EngMath is an ontology for mathematical modelling inengineering, written in Ontolingua[15]. It provides concep-tual foundations for representing mathematical and physi-cal entities such as scalars, vectors, tensors, physicalquantities, physical dimensions and units explicitly designedfor knowledge sharing applications in engineering [11].Regarding the unit representation problem, the mainfeature in EngMath (absent from MUO) is the compo-nent “physical dimensions”. The physical dimension of aquantity is an abstraction of a quantity ignoring magni-tude, sign and direction aspects[13]. The dimension of aquantity can be thought of as independent set of basedimensions[13]. For instance the quantity Body MassIndex (BMI) has the dimension that can be decomposedinto base dimensions mass and length: ML−2. The basedimensions in SI systems are length (L), mass (M), time(T), electric current (I), temperature (K), amount of sub-stance (N) and luminous intensity (J).EngMath (as opposed to MUO and UO) provides theenough semantic information to convert many unit-pairs of the same dimension that are either defined asbasic units or composed from the basic units[16]. Thekey limitation of EngMath for our purposes is that it isnot available in OWL. This problem is addressed in twomore recent ontologies QUDT and OM that use a simi-lar conceptual framework.QUDT: QUDT defines “quantity dimensions” whichallows for automatic consistency checking of differentquantities. QUDT also includes several major unit-sys-tems such as the CSG system of Units, SI and others[17], but relates all other unit systems back to SI usingtwo data properties - “conversion offset“, “conversionmultiplier“ - that could enable automated conversionbetween any non-SI-based unit and its SI-based equiva-lent. In terms of coverage for base units QUDT is fairlycomprehensive; however it lacks a number of derivedunits (e.g., the centimeters of mercury column com-monly used for clinical measurements of blood pres-sure); however it provides the framework within whichthese units could be created.OM: OM and QUDT are similar in terms of high leveldesign features and hence we only discuss the keySamadian et al. BMC Medical Genomics 2014, 7(Suppl 1):S12http://www.biomedcentral.com/1755-8794/7/S1/S12Page 2 of 8differences. One notable difference, is that OM defines“Quantity kinds” (e.g., acceleration and absorption) asOWL classes, which facilitates logical reasoning [13].Another advantage of OM -over QUDT- is that it uses theSI prefixes to relate submultiples (e.g., deci) and multiples(e.g., deca) of units. Additionally, OM defines the relation-ship between compound units (e.g., kilogram per cubicmeter) and their individual constituents (kilogram andcubic meter). Moreover, “compound units” is furtherdivided into “unit division” (e.g., meter per second), “unitexponentiation” (e.g., meter squared) and “unit multiplica-tion” (e.g., meter kilogram). For instance, for the unit“millimole per cubic centimeter” (mmol/cm3), the nomi-nator and denominator are defined as “millimole” and“cubic centimeter”, respectively, where millimole is relatedto mole by Prefix milli (om:factor = 1e-3) while “cubiccentimeter” is an instance of “unit exponentiation”. Thisadditional feature in OM allows for the automated calcula-tions of dimensions for new units in based on their consti-tuents. Figure 1 shows the model for “cubic centimeter”,revealing the relationship between OM classes that enableautomated reasoning and unit inter-conversion.In addition to above advantages, OM provides WebServices that can be used programmatically to incorpo-rate knowledge in OM in other applications making itsuitable for our study.MethodsDataset and data collectionThe dataset used for these investigations included the clin-ical records of a cardiovascular patient cohort collectedfrom a referral hospital in Nebraska, USA, between 1986and 1989, including 536 unique patients. Table 1 showssome columns from two rows of the dataset used in thisstudy. The intended meaning of acronyms for each col-umn header (e.g., SBP for Systolic Blood Pressure) wasconfirmed with the clinician who had originally collectedthe dataset. In the last four columns “1” represents “highrisk” and “0” represents a “low risk” for the condition listedin the header.As shown in the table, this dataset exhibited severalfeatures that presented significant challenges to auto-mated integration and annotation and that are commonwith legacy (and even contemporary) clinical data,including:1) The measurement-units were not representedexplicitly.2) Different rows of the same data set were repre-sented in different measurement units shown in dif-ferent colors. For instance, HDL is represented inmilligram/deciliter (Italic font) in the first row andin mmol/liter (Italic font) in the second row shownin different colors)3) The system of units used, even in the same row,could change (e.g., height in SI and weight in Imper-ial units in the first row).These observations highlight the need for a practicalsolution to ameliorate the problem of measurement-unitconflict resolution in health care.Data transformationOntologies and standards usedOM: As stated, OM provides a rich conceptualization ofthe compound units common in clinical data, weselected OM as our preferred unit-ontology startingpoint.GALEN: GALEN [18] is a rich compositional ontologyof the medical domain, covering anatomy, function, dis-eases, symptoms, drugs, and procedures. Following anapproach published previously[2] we re-factored andFigure 1 OM representation of cubic centimeter. The “compound units” in OM are divided into three top categories: “unit division”, “unitmultiplication” and “unit exponentiation” which provides additional information for automatically calculating the dimensions of new units.Samadian et al. BMC Medical Genomics 2014, 7(Suppl 1):S12http://www.biomedcentral.com/1755-8794/7/S1/S12Page 3 of 8extended a number of cardiovascular-relevant classes ofGALEN such that they could be used for logical reason-ing and classification.SADI and SHARE: SADI[19] is a set of standards-compliant design principles for exposing stateless WebServices on the Semantic Web. SADI Services consumeand produce RDF data, where the input and output dataproperties are described by OWL classes. These classesare, similarly, utilized to discover Services of interestthrough their registration in the SADI Service registry.SHARE[19] is an enhanced SPARQL query enginewhich is capable of (a) decomposing OWL classes intotheir constituent property restrictions, and (b) discover-ing and invoking SADI Services based on the propertiesthose Services consume/produce.SIO: The SemanticScience Integrated Ontology (SIO)is an ontology that provides models for the representa-tion of the scientific data [20]. It includes design princi-ples that facilitate creation of flexible software and isextensively used by analytical tools exposed using SADISemantic Web Services. Therefore our adoption of SIOin this work allows us to more easily take advantage ofexisting resources published using the SADI design pat-tern, as well as rapidly publish and integrate new toolsas-needed.Clinical data modelingThe GALEN ontology does not define the structure andproperties of individuals who would be members of itsontological classes, and therefore cannot easily be usedto classify data. As such, we extended the concepts inGALEN using SIO and OM. For example, we extendthe concept SystolicBloodPressure in GALEN as inOWL as follows:measure:SystolicBloodPressure =galen:SystolicBloodPressure and(“sio:has measurement value” some“sio:measurement” and(“sio:has unit” some “om:unit ofmeasure”) and(“om:dimension” value “om:pres-sure dimension”) and“sio:has value” some rdfs:Literal))In the above, the measurement units are linked toOM’s “pressure dimension”. This guarantees dimensioncompatibility during logical reasoning; i.e., all pressuredata in the clinical dataset are associated to “pressuredimension” by om:dimension relationship, and cantherefore automatically be directed to pressure-dimen-sion-relevant conversion services. Figure 2 shows theschematic view of the data model for systolic bloodpressure. The model provides a machine-processablemechanism for expressing semantics of measurement-units in the clinical domain.We followed OM’s framework for defining new unitsin order to create several unit-types commonly used inclinical science, but missing from OM’s existing set. Forinstance the unit “centimeter_of_mercury_column“(cmHg) which is often used for monitoring blood pres-sure does not exist in OM and was defined according toOM’s standards.Semantic Web servicesA single generic SADI-compliant Web Service was con-structed to manage conversions over the major quanti-ties (dimensions) of measurement most frequently usedin clinical setting including Pressure, Concentration,Table 1 Snapshot of the Dataset.ID HEIGHT WEIGHT SBP CHOL HDL BMIGRSBPGRCHOLGRHDLGRpt1 1.82 177 128 227 55 0 0 1 0pt2 179 196 13.4 5.9 1.7 1 0 1 0The first two rows of the dataset used in the original format. In the last fourcolumns “1” represents “high risk” and “0” represents a “low risk” for thecondition listed in the headerFigure 2 Extending clinical concepts. Extending clinical concepts in GALEN with richer logic including measurement values and units. Onceextended, Galen classes can be used for semantically enriched analyses.Samadian et al. BMC Medical Genomics 2014, 7(Suppl 1):S12http://www.biomedcentral.com/1755-8794/7/S1/S12Page 4 of 8Temperature, Length, and Mass. We first discuss thegeneric signature of the service and then describe anexample use cases for pressure and concentration (withminor differences) in more details. The other dimen-sions follow similar principles.The SADI service uses the semantics of the input datait receives to automatically configure itself to the correcttype of conversion. The input and output of this serviceare as follows:Input:“sio:has measurement value” some“sio:has unit” some “om: Unit of measure”and “sio:has value” some rdfs:LiteralOutput:“sio:has measurement value” some“sio:has unit” some “om: Unit of measure”and “sio:has value” some rdfs:Literaland “om:has dimension” exactly 1 om:DimensionThe following algorithmic steps are carried out auto-matically for each input data that meet the requirement:1. Find the dimension of input data. If input data isnot explicitly annotated with dimension, use the webservice provided by OM to annotate the dimensionof the input data unit.2. Use the dimension information to automaticallyconfigure the service to the appropriate conversion(e.g., Pressure convertor service).3. Use the dimension information as the input tofind all the compatible units (units with the samedimension)4. Iterate through compatible units frequently usedin clinical settings (e.g., iterate through all units forPressure) and apply the conversion for each (the lim-itations of this approach are discussed below)5. Use the API to calculate “unit conversion offset”(-getConversionOffset) and “unit conversion factor”(-getConversionFactor) to convert the input data intoselected compatible units. Attach the dimension andunit/value pairs to the output.ResultsPressureAll patient data was converted into RDF format using apattern similar to that shown in Figure 2. We thendefined an ontological class “High-Systolic-Blood-Pres-sure-Measurement”[21], as follows:measure:SystolicBloodPressure andsio:hasMeasurement some(sio:Measurement and ("sio:has unit”value om:kilopascal) and(sio:hasValue some double[>=“18.7"^^double])))This model uses units (kilopascals) that differ fromthose in our dataset (mmHg and cmHg), which allowsus to demonstrate the ability of the system to automati-cally detect and resolve unit conflicts. A SPARQL queryis provided to the SHARE query client that searches forhigh blood pressure measurements, as follows:SELECT ?record ?convertedvalue ?riskgradeFROM <./patient.rdf> WHERE{?record rdf:type measure:HighSystolicBloodPressure.?record sio:hasMeasurement ?measurement.?measurement sio:hasValue ?convertedvalue.?record cardio:ExpertClassification ?riskgrade.}In the above query SHARE examines the HighSystolic-BloodPressure class and discovers the “sio:has unit”value om:kilopascal axiom, indicating that measure-ments of High Systolic Blood Pressure will need to beexpressed in kilopascals. It compares this to the mea-surements in its dataset, and logically determines thatthey are discrepant in their units. It then queries theSADI registry to find a Service (the single Servicedescribed above) that provides conversions on the OMdimension of “Pressure” - i.e., a unit-conversion servicewith the input property of some “om:pressure dimen-sion”. It passes the measurement data to that service forunit-homogenization. For each individual incomingmeasurement, the unit is examined and the offset andcoefficient parameters required to convert betweensource (input data) and target (required by the OWLclass) unit are dynamically retrieved by the service usingthe OM REST interface. The conversion calculation isthen synthesized and applied to the incoming data.Once all data have been processed, the result is a data-set with all pressure units harmonized as kilopascals.The resulting output data is integrated into the localknowledge-base, prior to undertaking logical reasoningusing the Pellet reasoner, that classifies the patient dataas being consistent or inconsistent with the HighSysto-licBloodPressure ontological definition.Table 2 shows a demonstrative sub-set of the result datacorresponding to HighSystolicBloodPressure individuals,the showing a variety of pressure units being homogenizedTable 2 Units and values.RecordID StartValStartUnitEndValEnd Unit Expert’sclassificationcm_hg1 15 cmHg 19.998 kilopascal Highcm_hg2 14.6 cmHg 19.465 kilopascal Highmm_hg1 148 mmHg 19.731 kilopascal Highmm_hg2 146 mmHg 19.465 kilopascal HighUnits and valuse before and after conversion (2 digit precision).Samadian et al. BMC Medical Genomics 2014, 7(Suppl 1):S12http://www.biomedcentral.com/1755-8794/7/S1/S12Page 5 of 8the kilopascals in accordance with the OWL class defini-tion. Moreover, the system was able to identify all 134individuals, classified by the expert as having high systolicblood pressure, with no false positives. It should be notedthat though we did not find any misclassification, the pos-sibility of rounding errors introduced as a result of conver-sion cannot be ruled out; thus in future iterations it maybe desirable to specifically engineer the conversion servicesto make “sensible” choices about rounding, based onadvice from expert clinicians.ConcentrationThe term concentration most frequently refers toamount of substance in a solution. There are two majortypes of units that are used to represent concentration inclinical settings usually used to denote the concentrationof different chemicals in plasma (e.g., Hemoglobin). Thefirst type represents the amount of substance per unitvolume (e.g., gram per liter) and the second type repre-sents the number of moles per unit volume (e.g., moleper liter). Thus, we should note that, for concentration,both dimensions of mole/m3 (NL−3) and kg/m3 (ML−3)are used in practice. Molar-based units are routinely usedin medicine and physiology. As a result 1) the “conver-sion factor” between these different dimensions is not adimensionless parameter and 2) the conversion factorbetween depends on the molar mass of the specific mole-cule for which the measurement has been made. Forexample to convert mmol/L to mg/dL for Triglycerideand HDL we need to multiply by 88.57 and 38.67, respec-tively. A comprehensive list of molar-based units forplasma concentration of different chemicals and theircorresponding conversion factor can be found in [22].To achieve conversions in this case, it is necessary toknow the specific type of incoming measurement, basedon its GALEN class; for example, if it is a Triglyceridemeasurement, the incoming measurement must be ofrdf:type galen:tryglyceride, since the incoming unitscarry insufficient semantics in-and-of themselves to gen-erate the calculation parameters. Once the relevantcompound is determined from the logical type of theincoming measurement, the conversion constants arecalculated using the molecular mass of each molecule asper the look-up table in [22]. Here we only implementedthe concentration molecule-types used most frequentlyin clinical sciences (see supplementary materials) forperformance reasons; however the framework is extensi-ble by expanding the look-up table and the galenmolecule type ontology.Evaluations similar to those described for SystolicBlood Pressure were carried-out for Cholesterol andHDL classification, and the system was able to correctlyconvert and classify all records.Complex risk classification (revisited)As mentioned, in our previous study[2], we demon-strated that the combination of semantically-explicitdata, logically rigorous models of clinical guidelines, andpublicly-accessible Semantic Web Services, can be usedto execute automated, rigorous and reproducible clinicalclassifications. However, in the previous study our unitconversion services were written separately for eachquantity (one service for Pressure one service forConcentration and so on). In addition, the conversionformulas for conversion between different units werehard-coded within the Web Services. Additionally, usingour previous framework the relation between a quantityand its dimension was coded manually. In the currentexperiment, we addressed these shortcomings andrepeated the previous experiment. Not surprisingly, theresults for binary classifications were consistent withprevious experiment (see supplementary materials). Inthe following we will work through classification ofpatients based on their BMI value. The other risk classi-fications follow a similar pattern and are presented inthe supplementary materials.Modeling BMI risks is more complex than modellingconcentration, since BMI is derived by algorithmic ana-lysis of other core measurements (Height and Weight).BMI is calculated using a person’s weight and height.BMI in SI is calculated using the following formula.BMI =mass(kg)(height (m))2=mass (lb)(height (in))2× 703Once again, for evaluation purposes we intentionallydefined BMI in a different unit than the one presented inthe dataset. Due to reasons not discussed here (see [2]), theclinical researcher had used a different cut-off thresholdthan the ones suggested by American Heart Association(AHA) (26 kg/m2 was used instead of 25 kg/m2 to classifypatient as “Overweight”).Overweight =Patient and(sio:hasAttribute some(measure:BodyMassIndex andsio:hasMeasurement some (sio:Measurementand(sio:hasUnit value om:pound-per-inch- squared) and(sio:hasValue some double[>=18278.0]))))Where as before measure:BMI is extended using theBMI class in the GALEN ontology and 18278 (the pro-duction of 26 and 703) cut-off threshold is used to stra-tify patients. Subsequently, a SPARQL query wascomposed to classify records based on their BMI value.Samadian et al. BMC Medical Genomics 2014, 7(Suppl 1):S12http://www.biomedcentral.com/1755-8794/7/S1/S12Page 6 of 8SELECT ?record ?convertedvalue ?riskgradeFROM <./patient.rdf> WHERE{?record rdf:type measure:Overweight.?record sio:hasMeasurement ?measurement.?measurement sio:hasValue ?convertedvalue.?record cardio:ExpertClassification ?riskgrade.}It should be noted that BMI value does not exist inthe data prior to running the above query and it is cal-culated using the Semantic Web Service schematicallyshown in Figure 3 (sample data, and instructions onhow to send this data to the SADI service, are providedin the supplementary materials)Once we issue the SPARQL query above, the prop-erty-restriction imposed on the output, when detectedby SHARE, triggers the discovery and invocation of theService that attaches the BMI class with appropriateunits and value properties attached to it (first the servicecalculates all the possible output units and then itselects the one required by the output; see below forlimitations of the approach). Similar our previousexperiment, using this approach we were able to buildontological models that mirror the expert’s classificationof patients (based on BMI) of the individual clinicalresearcher with 100% accuracy.Limitations of the approachThe approach described in the examples above couldreasonably be criticized as being wasteful and/or ineffi-cient, since the unit-conversion Service converts anyincoming unit into all possible output units (limited tothose defined in a look-up table of the most commonunits used in clinical practice). This can cause perfor-mance issues; for instance a typical query takes on aver-age 35 minutes to resolve on a single machine (4 core3GHz machine with 8GB of memory) for the entirecohort. The problem is likely to exacerbate when dealingwith large and complex and distributed datasets. Thus,the current performance of the system is not acceptablefor a system designed to provide (close to) real-time sup-port. We can envision several ways by which perfor-mance improvement can be achieved. First, we note thatthe apparent wastefulness explained above is not necessa-rily a limitation of our Semantic approach to unit repre-sentation, but rather a limitation of the SHARE client’sinteraction with the unit-conversion Service. The Serviceis capable of receiving a “desired unit” parameter duringits invocation and, if present, it will use this informationto configure itself to do conversions non-wastefully; con-verting only into that specific unit, rather than all possi-ble units. The SHARE application, however, is notcapable of passing configuration parameters to a Serviceduring service invocation. Therefore, the apparent waste-fulness of the computations is an artefact of our use ofSHARE. Other SADI clients are currently under develop-ment, but were not available for this study. We reason-ably anticipate that by using such clients, significantperformance improvement can be achieved.There are several other areas other areas where signifi-cant improvements could be made. This includes differ-ent strategies for SPARQL query optimizations such asparallel (as opposed to sequential) processing of differentservices where possible, and development of strategies toavoid invocation of irrelevant services, by checking theinput and output signature (currently all the services thatFigure 3 Schematic diagram of the SADI Web Service. SADI web service interface to the BMI calculation Service. The property-restrictionimposed on the output, when detected by SHARE, triggers the discovery and invocation of the Service that attaches the BMI class withappropriate units and value properties attached to it.Samadian et al. BMC Medical Genomics 2014, 7(Suppl 1):S12http://www.biomedcentral.com/1755-8794/7/S1/S12Page 7 of 8attach a certain property will be invoked regardless of theinput and output datatype) of the services.ConclusionsUnit conversion is a common and troublesome barrier tointegration. Busy health researchers should not need toconcern themselves this trivial, error-prone, but neces-sary exercise. Here we have utilized a combination ofsemantic standards and frameworks to demonstrate, withseveral unit-conversion exemplar cases, that these typesof data integration problems can, and should, be dealtwith by the machines themselves. By encoding data withsemantic transparency, it becomes possible for machinesto detect unit conflicts and use semantic systems such asSADI + SHARE to automatically resolve them.We note that a large number of measurement units inclinical practice include more complex unit patternsthan the ones we modeled. For instance the majority ofunits used for drug dosage and clearance include tem-poral elements (e.g., mg/kg/hour for the drug dosage)that are not modeled in this study. To the best of ourknowledge such patterns have not been modeled in anyexisting ontologies. As such, we plan to extend our fra-mework to include more complex patterns togetherwith temporal units and their conversion. Finally, weplan to extend our study to include different datasetsfrom multiple centers and evaluate the usability of ourapproach in more complex biomedical scenarios.Additional informationSupplementary materials, and working examples, areprovided at: http://biordf.org/MeasurementUnitsDemo.Competing interestsThe authors declare that they have no competing interests.Authors’ contributionsSS and MW planned this work, and jointly wrote this manuscript. SSexecuted the data migration, ontology design and extension, web servicedeployment, and overall analysis. BM generated and initially analyzed thesource clinical dataset, and discussed and validated our approach andchoice of clinical standards. All authors have read, revised and approved thismanuscript.DeclarationsPublication of this article has been funded by Special Initiatives Grant-in aidfrom the Heart and Stroke Foundation of British Columbia and Yukon,Microsoft Research and an Operating Grant from the Canadian Institutes ofHealth Research (CIHR). MDW is funded by the Isaac Peral/Marie Curie co-fund program at the Technical University of Madrid.This article has been published as part of BMC Medical Genomics Volume 7Supplement 1, 2014: Selected articles from the 3rd TranslationalBioinformatics Conference (TBC/ISCB-Asia 2013). The full contents of thesupplement are available online at http://www.biomedcentral.com/bmcmedgenomics/supplements/7/S1.Authors’ details1UBC James Hogg Research Center, Institute for Heart + Lung Health, Room166 - 1081 Burrard Street, St. Paul’s Hospital Vancouver, BC, Canada, V6Z 1Y6.2Centro de Biotecnología y Genómica de Plantas, Universidad Politécnica deMadrid, Madrid, España.Published: 8 May 2014References1. Welch DL: Human error and human factors engineering in health care.Biomedical Instrumentation and Technology 1997, 31:627-31.2. Samadian S, McManus B, Wilkinson MD: Extending and encoding existingbiological terminologies and datasets for use in the reasoned semanticweb. Journal of Biomedical Semantics 2012, 3:6.3. Scheider S, Janowicz K, Kuhn W: Grounding geographic categories in themeaningful environment. Spatial Information Theory 2009, 2009:69-87.4. Mari L: Epistemology of measurement. Measurement 2003, 34:17-30.5. Frigerio A, Giordani A, Mari L: Outline of a general model ofmeasurement. Synthese 2009, 175:123-149.6. Gkoutos GV, Schofield PN, Hoehndorf R: The Units Ontology: a tool forintegrating units of measurement in science. The Journal of BiologicalDatabases and Curation 2012.7. Schadow G, McDonald CJ, Suico JG, Föhring U, Tolxdorff T: Units ofmeasure in clinical information systems. Journal of American MedicalInformatics Associacion 6(2):151-62.8. RDF - Semantic Web Standards. [http://www.w3.org/RDF/].9. OWL Web Ontology Language Overview. [http://www.w3.org/TR/owl-features/].10. MUO: Measurement Units Ontology. [http://idi.fundacionctic.org/muo/muo-vocab.html].11. RUBER TG, LSEN GO: An ontology for engineering mathematics.Proceedings of Fourth International Conference on Principles of KnowledgeRepresentation and Reasoning 1994, 258-269.12. QUDT - Quantities, Units, Dimensions and Types. [http://www.qudt.org/].13. Ontology of Units of Measure and Related Concepts. [http://www.semantic-web-journal.net/content/ontology-units-measure-and-related-concepts].14. Samadian S: Constructing and Applying Semantic Models of ClinicalPhenotypes to Support Web-embedded Clinical Research. PhD Thesis2013.15. Ontolingua Home Page. [http://www.ksl.stanford.edu/software/ontolingua/].16. Ontolingua Theory STANDARD-UNITS. [http://www-ksl.stanford.edu/knowledge-sharing/ontologies/html/standard-units/index.html].17. Cardarelli FO: Encyclopaedia of Scientific Units, Weights and Measures:Their Si Equivalences and Origins. Springer;, 1 2003, 848.18. Rector A, Rogers J, Pole P: The GALEN High Level Ontology. Proceedings ofMIE 96 1996, 174-178.19. Wilkinson MD, Vandervalk B, McCarthy L: The Semantic AutomatedDiscovery and Integration (SADI) Web service Design-Pattern, API andReference Implementation. Journal of Biomedical Semantics 2011, 2:8.20. Vandervalk B: The SHARE System: A Semantic Web Based Approach forEvaluating Queries Across Distributed Bioinformatics Databases andSoftware. MSc Thesis 2011.21. Chobanian AV, Bakris GL, Black HR, Cushman WC, Green LA, Izzo JL,Jones DW, Materson BJ, Oparil S, Wright JT, Roccella EJ: The SeventhReport of the Joint National Committee on Prevention, Detection,Evaluation, and Treatment of High Blood Pressure: the JNC 7 report.JAMA 2003, 289:2560-72.22. International System of Units. [http://www.questdiagnostics.com/dms/Documents/test-center/si_units.pdf].doi:10.1186/1755-8794-7-S1-S12Cite this article as: Samadian et al.: Automatic detection and resolutionof measurement-unit conflicts in aggregated data. BMC MedicalGenomics 2014 7(Suppl 1):S12.Samadian et al. BMC Medical Genomics 2014, 7(Suppl 1):S12http://www.biomedcentral.com/1755-8794/7/S1/S12Page 8 of 8

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.52383.1-0223374/manifest

Comment

Related Items