Toward Pay-As-You-Go Data Integration for Healthcare Simulations

Philipp Baumg

artel

1,2

, Gregor Endler

and Richard Lenz

Institute of Computer Science 6, Friedrich-Alexander University of Erlangen-Nuremberg, Erlangen, Germany

On behalf of the ProHTA Research Group

Keywords:

Data Management, Data Integration, Simulation, Healthcare.

Abstract:

ProHTA (Prospective Health Technology Assessment) aims at understanding the impact of innovative medical

processes and technologies at an early stage. To that end, large scale healthcare simulations are employed to

estimate the effects of potential innovations. Simulation techniques are also utilized to detect areas with a high

potential for improving the supply chain of healthcare. The data needed for both validating and adjusting these

simulations typically comes from various heterogeneous sources and is often preaggregated and insufﬁciently

documented. Thus, new data management techniques are required to cope with these conditions. Because of

the high initial integration effort, we propose a pay-as-you-go approach using RDF. Thereby, data storage is

separated from semantic annotation. Our proposed system offers automatic initial integration of various data

sources. Additionally, it provides methods for searching semantically annotated data and for loading it into

the simulation. The user can add annotations to the data in order to enable semantic integration on demand.

In this paper, we demonstrate the feasibility of this approach with a prototype implementation. We discuss

beneﬁts and remaining challenges.

1 INTRODUCTION

Medical and statistical simulation data in our project

ProHTA (Prospective Health Technology Assess-

ment)(Djanatliev et al., 2012) stems from several

heterogeneous sources like spreadsheets, relational

databases or XML ﬁles. Therefore, semantic data in-

tegration is an important concern. Although many

techniques exist for automatic integration of data

(Rahm and Bernstein, 2001), it is still an expensive

process. Like Lenz et al. (Lenz et al., 2007), we dis-

tinguish between technical and semantic integration.

Technical integration enables accessing data. In con-

trast, semantic integration facilitates understanding

the meaning of data. Since our pool of data sources

is likely to change frequently, we investigate an in-

tegration strategy based on the dataspaces abstraction

(Franklin et al., 2005). This approach allows the coex-

istence of heterogeneous data sources, which are ini-

tially integrated only as far as automatically possible

(Das Sarma et al., 2008). The system then allows the

gradual improvement of the degree of integration in a

pay-as-you-go manner (Jeffery et al., 2008).

To achieve pay-as-you-go integration, we devel-

oped an architecture with multiple levels of integra-

tion based on the ﬁve-level schema architecture for

federated databases by Sheth and Larson (Sheth and

Larson, 1990). The original ANSI/SPARC three-

level schema architecture consists of conceptual, in-

ternal, and external schema. The conceptual schema

describes the data structures and the relationships

among them on a logical level. The internal schema

contains indexes and information about the physical

storage of records. The external schema provides tai-

lored views for separate groups of users. Sheth and

Larson (Sheth and Larson, 1990) introduce additional

schema levels to support federated databases. The lo-

cal schema contains the native data model of a data

source. The component schema is the translation of

the local schema into a canonical data model. For our

approach, we adapt the local schema and component

schema, which replace the conceptual schema of the

ANSI/SPARC architecture. By explicitly storing the

native data model of a data source and deferring the

translation into a canonical data model, we allow for

demand driven integration.

To enable storing heterogeneous data with differ-

ent data models, our data management system em-

ploys a fully generic data schema. As a side effect,

following the principle of “design for change” (Par-

nas, 1994) from the outset ensures evolutionary capa-

bilities, an important factor for success. This is espe-

172

Baumgärtel P., Endler G. and Lenz R..

Toward Pay-As-You-Go Data Integration for Healthcare Simulations.

DOI: 10.5220/0004734201720177

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2014), pages 172-177

ISBN: 978-989-758-010-9

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

cially true for projects in the healthcare sector with its

rapidly changing conditions (Lenz, 2009).

An evolutionary information system requires spe-

cial attention as to the keeping of data. A fully

generic data schema like EAV/CR (Nadkarni et al.,

1999), while perfectly able to manage changing

conditions, signiﬁcantly increases query complexity.

For this reason, we adopt an approach using RDF

(Resource Description Framework) (Lassila et al.,

1999). RDF employs a generic schema consisting of

(subject,predicate,object) triples to store and

link objects. In (Baumg

artel et al., 2013), we de-

veloped a Benchmark to compare the performance

of several generic data management solutions for

simulation input data. With this benchmark, we

evaluated a document store, relational databases us-

ing an EAV schema, and an RDF triplestore. This

benchmark showed that RDF triplestores are slower

and not as mature as relational databases. How-

ever, RDF triplestores allow concise SPARQL queries

(Prud’hommeaux and Seaborne, 2008), whereas SQL

queries on an EAV/CR schema prove to be very com-

plex. Therefore, we chose to use RDF triplestores, as

their ﬂexibility and query capabilities outweigh their

performance drawbacks.

Howe et al. (Howe et al., 2011) identiﬁed

several key barriers to adopting a data management

system for science. They argue that the initial effort

of designing a data schema is too complicated in

most cases. Therefore, a “data ﬁrst, structure later”

approach should be implemented to reduce the

initial effort. Another problem identiﬁed by Howe

et al. is that scientists sometimes need help to write

non-trivial SQL statements.

Therefore, we identiﬁed several requirements for

our simulation data management system:

Automatic Technical Integration: Because of

the changing demands of the simulation and the

amounts of heterogeneous data, the initial effort of

integrating a data source should be minimized. Many

sources being available to us have the same format

but different semantics. Therefore, the system should

enable automatic technical integration and schema

import for sources with known formats.

Deferred Semantic Integration: As the data has

to be reusable for different simulation studies, the

data management system has to provide the means of

adding semantic information to stored data. These an-

notations can then be used for deferred semantic inte-

gration on demand implementing a “data ﬁrst, struc-

ture later” approach.

Querying Simulation Data: The simulation

modeler has to be able to ﬁnd and query data easily.

The effort of using a data provider for simulation

input data management should be less than manual

data input.

2 INTEGRATION

In this section, we propose an approach to simulation

data management utilizing pay-as-you-go integration.

In a previous paper (Baumg

artel and Lenz, 2012),

we developed an ontology for storing data cubes us-

ing RDF. We also described a simple domain speciﬁc

query language assisting simulation modelers to load

data into their simulation models. However, the prob-

lem of high initial effort for integrating the data re-

mained unsolved, as up-front semantic information

about the data sets is necessary.

Therefore, we propose a ﬂexible solution for in-

tegrating, annotating and querying simulation input

data. Data and metadata are stored in an RDF triple-

store using several generic upper ontologies. A web

frontend visualizes the data and provides methods for

getting user input. Figure 1 depicts the components of

our approach. Data is loaded using input adapters for

various source types, e.g. spreadsheets, XML ﬁles or

relational databases. Then, the data is stored utilizing

the automatically imported schema of the data source.

For example, a spreadsheet is stored using concepts

like table, row, column and cell. Initially, the user can

query the data utilizing full-text search or SPARQL.

Also, the user is able to view the data employing a vi-

sualization module for that speciﬁc source type. This

module also enables the user to add annotations about

the semantics of the data to facilitate deferred seman-

tic integration.

Figure 1: The architecture of our simulation data manage-

ment system.

TowardPay-As-You-GoDataIntegrationforHealthcareSimulations

173

Meta Information

Data Structures

Annotations

AnnotationClass

AnnotationSetClass

elementOfClass

Annotation

DataClass

fromClass toClass

AnnotationSet

Data

DataStructure

subClassOf

DataItem

subClassOf

Literal

value

Element

contains

elementOf

hasAnnotation

toDataStructure

elementOf

Literal

value

Caption

RDF Property with Domain and Range

RDF Tripel

Figure 2: Upper ontology for data structures and annotations.

2.1 Example

As a running example, we use the integration of mul-

tidimensional statistical data. This kind of data is the

most prominent in our simulation project. We have to

integrate preaggregated statistical data in pivot tables

stored in spreadsheets. Figure 3 shows an example for

this kind of data.

To integrate pivot tables as multidimensional data

cubes, we need semantic information about the di-

mensions represented by rows and columns. In the

following sections, we will show how our generic

framework for demand driven integration enables us

to build a pay-as-you-go integration engine to accom-

plish this.

Figure 3: Our web frontend visualizing tabular data.

2.2 Technical Integration

For every type of data source an input adapter has to

be implemented and an ontology to store the schema

of this source type has to be constructed. We devel-

oped an upper ontology for describing these schema

ontologies. This ontology is depicted in Figure 2. Ad-

ditionally, this upper ontology enables storing arbi-

trary data that corresponds to a schema ontology. The

data is stored as Elements of Data Structures. These

Elements may contain Data, which could be a single

Data Item or another substructure. Therefore, sev-

eral Data Structure ontologies can be assembled. El-

ements can have Annotations that are part of a set of

annotations for one Data Structure.

Now, ontologies to store data cubes or tables can

be constructed using this upper ontology. To integrate

a data source, the input adapter imports the actual

schema of this source and stores it as an instance of a

Data Structure. Then, the data is imported and stored

in the RDF triplestore. This enables querying of all

data sources with one common query language. Us-

ing the terminology of Sheth and Larson (Sheth and

Larson, 1990), the schema of the data source corre-

sponds to the local schema.

To validate this approach, we implemented an

adapter for CSV ﬁles and an adapter for spreadsheets,

as these are the most common ﬁle types in our project.

To store these ﬁles, we constructed an ontology for

tabular data using our upper ontology (Figure 4(a)).

The input adapter uses this ontology to import the

schema of the data sources. All of the classes in the

table ontology are derived from classes of our upper

ontology. As all data is stored using our upper ontol-

ogy as a common vocabulary, our system can initially

provide a full-text keyword search for these ﬁle types.

Additionally, the data can be queried using SPARQL.

However, our approach is not limited to spreadsheets

but also applicable to relational or hierarchical data

for example.

HEALTHINF2014-InternationalConferenceonHealthInformatics

174

2.3 Visualization and Semantic

Annotation

For every source type, a visualization module can be

implemented. To display the data, we use HTML tem-

plates and modules to load and process schema infor-

mation. We implemented a visualization component

for tabular data to visualize CSV ﬁles and spread-

sheets. Fig. 3 depicts our web frontend displaying a

simple table.

The frontend facilitates the annotation of individ-

ual elements of a dataset to add semantic informa-

tion. To this end, the frontend automatically generates

annotation forms using annotation ontologies. These

annotation ontologies describe sets of possible anno-

tations. To deﬁne these annotation ontologies, we

developed an upper annotation ontology (Figure 2).

These semantic annotations can then be utilized for

semantic integration.

For tabular data, the elements that can be anno-

tated are rows, columns and cells. The user can se-

lect a set of possible semantic annotations he would

like to add. For example, the user can annotate rows

and columns of a spreadsheet containing multidimen-

sional data with information about dimensions and

granularities. Figure 4(b) depicts an annotation ontol-

ogy that can be used to store information about mul-

tidimensional data in tables.

To validate the modularity of our approach, we

implemented an additional annotation ontology to

describe spreadsheets containing multiple subtables.

For this new annotation ontology, the annotation

forms are generated automatically and the table can

be annotated. Because of the independence of the

modules, there is no need to change any code when

adding new annotation ontologies. However, mod-

ules to process the annotations for semantic integra-

tion have to be implemented.

2.4 Semantic Integration

When there is sufﬁcient semantic information about

a dataset, the system can translate the data into the

canonical data model, which is represented by another

data-structure ontology. For each set of annotations

and each source type, a semantic integration module

can be implemented.

We implemented a module to integrate mul-

tidimensional data using our data cube ontology

(Baumg

artel and Lenz, 2012) as the canonical data

model. This component utilizes annotations to ta-

bles (Figure 4(b)) and constructs new multidimen-

sional data structures. That way, we can integrate

pivot tables as multidimensional data cubes. Now, the

data can be queried using our domain speciﬁc query

language for multidimensional data (Baumg

artel and

Lenz, 2012). Other domain speciﬁc languages can be

added to our query processor as well.

2.5 Searching and Querying Simulation

Input Data

After data is stored in our system, it offers a full-

text keyword search. As SPARQL queries tend to be

very complex, we do not expect the simulation mod-

elers to write SPARQL queries themselves. In future

work, we will add automatic query generation to the

graphical user interface. However, writing SPARQL

queries remains possible and enables the user to ag-

gregate and manipulate the data. As data is annotated

with further information, the data can be queried us-

ing high level domain speciﬁc query languages. For

example, we developed a domain speciﬁc query lan-

guage for agent based simulation models to query and

aggregate data that corresponds to steps in a UML ac-

tivity diagram (Baumg

artel et al., 2014).

3 VALIDATION

We validated our concepts by integrating CSV-ﬁles

and spreadsheets stemming from the German Federal

Statistical Ofﬁce

. Most of these data ﬁles contained

multidimensional data as pivot tables.

We compared the traditional data manage-

ment workﬂow using spreadsheets or manual input

(Robertson and Perera, 2002) to the workﬂow using

our system. To evaluate the traditional workﬂow, we

employed a document management system to store

spreadsheets. In both scenarios, data is stored by up-

loading ﬁles to a web frontend. Both the document

management system and our prototype offer means of

searching data by keywords. In the traditional work-

ﬂow the data is viewed using a spreadsheet applica-

tion. In contrast, our system offers integrated means

of viewing spreadsheet data. In the traditional work-

ﬂow, data sets are loaded into the simulation model

by simulation framework speciﬁc adapters for spread-

sheets. Using our approach, data sets are loaded into

the simulation using SPARQL. The effort of deﬁning

the desired rows and columns of the data set using

the simulation framework speciﬁc adapters was equal

to using SPARQL. However, semantic annotation and

integration enable the use of domain speciﬁc query

languages with additional features. Therefore, the ini-

tial effort of storing data is the same in both scenar-

http://www.destatis.de

TowardPay-As-You-GoDataIntegrationforHealthcareSimulations

175

DataStructure

Table

subClassOf

Element

Row

subClassOf

Column

subClassOfCell

subClassOf

Integer

rowNr

Integer

columnNr

row column

(a) Ontology for tables

AnnotationSet

TableToCube

subClassOf

Table

fromClass

Cube

toClass

Dim

elementOfClass

Granularity

elementOfClass

Annotation

subClassOf subClassOf

Row

rowIsDim

Column

colIsDim rowInGranularity colInGranularity

RDF Property with

Domain and Range

RDF Tripel

(b) Ontology to annotate pivot tables

Figure 4: RDF classes to describe tables and the mapping of pivot tables to data cubes.

ios. However, our system offers techniques to gradu-

ally improve the semantic integration of the data and

therefore its reusability.

4 RELATED WORK

In this section, we will give a brief overview about

existing work on simulation data management.

SciDB (Rogers et al., 2010) is a tool for scientiﬁc

data management storing large multidimensional ar-

ray data and supporting efﬁcient data processing. In

contrast to SciDB, our main challenge is not to handle

very large array data, but to handle many small data

sets in heterogeneous formats.

SIMPL (Reimann et al., 2011) is a framework sup-

porting the ETL-process (extract, transform and load)

for simulation data. The schema of data sources is

described in XML, which is ﬂexible enough for het-

erogeneous data.

DaltOn (Jablonski et al., 2009) is a data integra-

tion framework for scientiﬁc applications. Integration

is done using descriptions of the data sources.

Arthofer et al. (Arthofer et al., 2012) devel-

oped an ontology based tool to integrate medical data.

They also provide an automatically generated ontol-

ogy based web frontend to add additional information

with focus on data quality.

However, with all these systems the high initial ef-

fort of data integration remains. Howe et al. (Howe

et al., 2011) describe a relational scientiﬁc dataspace

system with support for schema free storage of tables.

The main aspect of the system is the automatic sug-

gestion of SQL queries and not the semantic integra-

tion of the tables. In contrast, our system facilitates

semantic integration and can be extended to support

other data formats than tables.

5 CONCLUSIONS

In this paper, we proposed a data management system

for healthcare simulations. We clariﬁed the impor-

tance of a pay-as-you-go data integration approach for

simulation data management. After the discussion of

the basic concepts, we described our prototypical im-

plementation and the validation of our concepts. Fi-

nally, we discussed related work.

Our modular framework for heterogeneous data

and input adapters facilitates automatic initial tech-

nical integration. Additionally, visualization and an-

notations enable deferred semantic integration. This

“data ﬁrst, structure later” approach minimizes the

initial integration effort. Searching and querying

the data is possible utilizing a keyword search and

SPARQL. Finally, semantic integration enables en-

hanced query possibilities like domain speciﬁc query

languages. We validated our approach with a proto-

typical implementation of our framework.

In future work, we will improve our framework by

adding support for various other types of data sources

and improving usability. Additionally, we will add

facilities for automatic query generation to our fron-

tend to support simulation modelers. Finally, we will

evaluate our approach in the context of our simulation

project.

ACKNOWLEDGEMENTS

This project is supported by the German Federal Min-

istry of Education and Research (BMBF), project

grant No. 13EX1013B.

HEALTHINF2014-InternationalConferenceonHealthInformatics

176

REFERENCES

Arthofer, K., Girardi, D., and Giretzlehner, M. (2012).

Ein ontologiebasiertes system zum extrahieren, trans-

formieren und laden in krankenanstalten. In Proceed-

ings of the eHealth2012.

Baumg

artel, P., Endler, G., and Lenz, R. (2013). A bench-

mark for multidimensional statistical data. In Catania,

B., Guerrini, G., and Pokorn, J., editors, Advances in

Databases and Information Systems, volume 8133 of

Lecture Notes in Computer Science, pages 358–371.

Springer Berlin Heidelberg.

Baumg

artel, P. and Lenz, R. (2012). Towards data and data

quality management for large scale healthcare simu-

lations. In Conchon, E., Correia, C., Fred, A., and

Gamboa, H., editors, Proceedings of the International

Conference on Health Informatics, pages 275–280.

SciTePress - Science and Technology Publications.

Baumg

artel, P., Tenschert, J., and Lenz, R. (2014). A query

language for workﬂow instance data. In Catania, B.,

Cerquitelli, T., Chiusano, S., Guerrini, G., Kmpf, M.,

Kemper, A., Novikov, B., Palpanas, T., Pokorn, J., and

Vakali, A., editors, New Trends in Databases and In-

formation Systems, volume 241 of Advances in Intel-

ligent Systems and Computing, pages 79–86. Springer

International Publishing.

Das Sarma, A., Dong, X., and Halevy, A. (2008). Boot-

strapping pay-as-you-go data integration systems. In

Proceedings of the 2008 ACM SIGMOD international

conference on Management of data, SIGMOD ’08,

pages 861–874, New York, NY, USA. ACM.

Djanatliev, A., Kolominsky-Rabas, P., Hofmann, B. M., and

German, R. (2012). Hybrid simulation with loosely

coupled system dynamics and agent-based models for

prospective health technology assessments. In Pro-

ceedings of the 2012 Winter Simulation Conference.

Franklin, M., Halevy, A., and Maier, D. (2005). From

databases to dataspaces: a new abstraction for infor-

mation management. SIGMOD Rec., 34:27–33.

Howe, B., Cole, G., Souroush, E., Koutris, P., Key, A.,

Khoussainova, N., and Battle, L. (2011). Database-

as-a-service for long-tail science. In Bayard Cush-

ing, J., French, J., and Bowers, S., editors, Scientiﬁc

and Statistical Database Management, volume 6809

of Lecture Notes in Computer Science, pages 480–

489. Springer Berlin / Heidelberg.

Jablonski, S., Volz, B., Rehman, M., Archner, O., and Cur,

O. (2009). Data integration with the dalton frame-

work - a case study. In Winslett, M., editor, Scientiﬁc

and Statistical Database Management, volume 5566

of Lecture Notes in Computer Science, pages 255–

263. Springer Berlin / Heidelberg.

Jeffery, S. R., Franklin, M. J., and Halevy, A. Y. (2008).

Pay-as-you-go user feedback for dataspace systems.

In Proceedings of the 2008 ACM SIGMOD interna-

tional conference on Management of data, SIGMOD

’08, pages 847–860, New York, NY, USA. ACM.

Lassila, O., Swick, R. R., Wide, W., and Consortium, W.

(1999). Resource description framework (rdf) model

and syntax speciﬁcation. http://www.w3.org/TR/

1999/REC-rdf-syntax-19990222.

Lenz, R. (2009). Information systems in healthcare - state

and steps towards sustainability. IMIA Yearbook 2009,

1:63–70.

Lenz, R., Beyer, M., and Kuhn, K. A. (2007). Semantic inte-

gration in healthcare networks. International Journal

of Medical Informatics, 76(2-3):201 – 207.

Nadkarni, P. M., Marenco, L., Chen, R., Skoufos, E., Shep-

herd, G., and Miller, P. (1999). Organization of het-

erogeneous scientiﬁc data using the eav/cr represen-

tation. Journal of the American Medical Informatics

Association, 6(6):478–493.

Parnas, D. L. (1994). Software aging. In Proceedings of the

16th international conference on Software engineer-

ing, ICSE ’94, pages 279–287, Los Alamitos, CA,

USA. IEEE Computer Society Press.

Prud’hommeaux, E. and Seaborne, A. (2008). Sparql query

language for rdf. http://www.w3.org/TR/2008/

REC-rdf-sparql-query-20080115/.

Rahm, E. and Bernstein, P. A. (2001). A survey of ap-

proaches to automatic schema matching. The VLDB

Journal, 10:334–350. 10.1007/s007780100057.

Reimann, P., Reiter, M., Schwarz, H., Karastoyanova, D.,

and Leymann, F. (2011). Simpl - a framework for

accessing external data in simulation workﬂows. In

Datenbanksysteme f

ur Business, Technologie und Web

(BTW).

Robertson, N. and Perera, T. (2002). Automated data collec-

tion for simulation? Simulation Practice and Theory,

9(6-8):349 – 364.

Rogers, J., Simakov, R., Soroush, E., Velikhov, P., Balazin-

ska, M., DeWitt, D., Heath, B., Maier, D., Madden,

S., Patel, J., Stonebraker, M., Zdonik, S., Smirnov, A.,

Knizhnik, K., and Brown, P. G. (2010). Overview of

scidb, large scale array storage, processing and analy-

sis. In Proceedings of the SIGMOD’10.

Sheth, A. P. and Larson, J. A. (1990). Federated database

systems for managing distributed, heterogeneous, and

autonomous databases. ACM Comput. Surv., 22:183–

236.

TowardPay-As-You-GoDataIntegrationforHealthcareSimulations

177