Description
Enabling Business Intelligence Functions over a Loosely Coupled Environment
Enabling Business Intelligence Functions over a Loosely
Coupled Environment
Giampaolo Armellin
1
, Leandro Paulo Bogoni
1,2
, Annamaria Chiasera
1,2
,
Tefo James Toai
1
, Gianpaolo Zanella
1
1
GPI SpA, Via Ragazzi del '99, 13 - Trento, Italy
2
Information Engineering and Computer Science Department - University of Trento,
Trento, Italy
[email protected],
{garmellin, achiasera, ttoai, gzanella}@gpi.it
Abstract. Planning effective and well targeted actions to manage and improve
the local and national healthcare services requires institutions to understand and
analyse the real needs of the population based on reliable and timely statistical
analysis on citizens’ health state. This is particularly important in developing
countries in which healthcare facilities lack ICT infrastructures and network
connectivity, making data collection and analysis particularly difficult with a
considerable manual effort leading to potentially unreliable or incoherent
information. In this scenario, we propose a generic communication
infrastructure, developed in the SIS-H project for Mozambique hospitals to
capture, communicate and analyse clinical events. Our solution enables the
exchange of data amongst healthcare facilities over all the different aggregation
levels of the hierarchical healthcare system of Mozambique regardless of the
availability of communication media (e.g., compact disk, usb stick, web-
internet). The plugin-based solution adopted supports reporting and Business
Intelligence analysis for exploring data at different granularity levels.
Keywords: Business Intelligence, e-Governance, plugin-based architecture
1 Introduction
Improving accessibility and quality of health services is one of the outcomes of the
Mozambican Health Sector Strategic Plan (PESS 2007-2012 6, 5). Despite public and
external founds, medical assistance is not accessible to all the population owing to the
difficulty in coordinating and controlling the actions performed and a chronicle
deficiency of human resources. The World Health Organization points out the need of
monitoring systems to control resource flows, progresses and outcomes finalized to
monitor and evaluate health services [5], [9]. In the Strategic Plan for the Health
Sector [6] the role of such monitoring system is essential for decision makers to
manage and supervise the maintenance of hospitals. Another pre-requisite to evaluate
and improve health quality is to encourage community participation with the active
communication of the results of the analysis promoting transparency and
accountability.
In this paper we will report our experience in the development of a communication
infrastructure applicable at the national level for the monitoring of the quality of
healthcare system in Mozambique. We propose a lightweight and robust infrastructure
that requires minimum resources to operate with poor or no connectivity to support
people in the collection, processing and communication of statistical information with
a simple and intuitive tool usable even by not trained personnel.
1.1 The Context
The structure of the Mozambican Healthcare System is level-oriented and hierarchical
– i.e. operational units and hospitals are grouped according to managed services and
geographical areas. The central hospitals of Beira and Maputo are at the top levels of
the hierarchy (level IV hospitals) and refer directly to the ministry of health. Pemba
provincial hospital is at level III and refers to the Provincial Health Office.
The ministry of health and the provinces provide economical support to operational
units and hospitals and plan actions and projects to prevent and deal with chronicle
and highly contagious diseases (like Malaria, HIV/AIDS, Cholera). These activities
include, among others, economical investments to improve the infrastructures
(hospitals’ facilities like emergency rooms, intensive units, infective divisions), staff
training, and education of the population for prevention.
Governing bodies need to know which are the real needs of the population to
identify and deal with sanitary emergency, to plan an effective set of activities (e.g.
the continuous and periodic surveillance required by HIV epidemic [7]) and to reduce
waste of resources (staff and financial). In [10] it is highlighted how this requires at
first the surveillance of the morbidity and mortality causes. The analysis of such
statistical data allows on one side to understand the health state of the population and
on the other side to monitor hospital resource utilization and in particular costs of
hospitalization and staff employed. Based on these analyses the ministry of health can
better plan public health interventions, for example to grant a suitable number of beds
in hospitals or health posts facing particularly critical sanitary crisis.
As pointed out in the analysis of Campione et al. [10] the social, economical and
technological environment of Mozambique poses some strict constraints to realize
such analysis that in developed countries are normal Business Intelligence activities
[2]. In particular [10] identifies the following challenges:
? lack of IT skills: qualified personnel, able to operate IT devices and tools is scarce
and consequently the data collected is not trustworthy;
? lack of data: the diagnostic capability of the lower level point of care is rather poor
compared to more specialized level of care (hospitals of level III or below have
less diagnostic facilities than hospitals of level IV);
? territory highly distributed and sparsely populate: in order to get a comprehensive
picture of the health state of population it is necessary to have a good coverage of
the territory at the national level;
? infrastructure limitations: no electricity, no information communication
technologies, no connectivity and no computers available;
? lack of information systems: there are neither the applications to manage health
records nor the personal information of patients and consequently it is difficult to
identify patients in the long term and to maintain detailed information on their
health story.
Data is typically collected on paper with significant manual effort and difficulties
of communication and analysis by other parties at higher level of the hierarchy where
planning and resource management is performed. The lack of computerization in the
point of care, especially at the periphery, makes the data collection particularly
difficult; with a considerable amount of time and manual effort required from the
personnel. Furthermore, people are not motivated to spend effort in such a boring task
as they do not see any useful results in short time coming out. These factors leave the
data collection process incomplete and error-prone.
Indeed a great step forward has been performed with the adoption of a standardized
list of diagnosis selected from the ICD-10 standard [4]. However, an IT solution
capable of effectively supporting the collection of data, and is easy to install in
critical environment with lack of infrastructure and is simple to use even for poorly
trained people; is still missing. This solution should be general enough to be usable at
any level of the national health care system with a great degree of flexibility and
customizability to collect as much information as possible.
In this work we present the solution we provide in the context of the project
“SIS-H – Módulo Internamentos” launched with internal tender by the Mozambican
Ministry of Health (MISAU) according to the Mozambique eGovernment Strategy.
The main project goal is to rapidly overcome the lack of statistical information for
hospitals of level III and level IV, devising software applications to collect and
aggregate data in the clinical area, with specific focus on admissions and discharges
of patients. Those aggregated data concern Maputo Central Hospital, Beira Central
Hospital and Pemba Provincial Hospital and are being periodically sent to Provincial
Health Offices and to the Ministry of Health, enabling analysis and planning on the
health care system.
We provide a platform that effectively answers the challenges identified above
both technical (lack of IT infrastructure) and organizational (data sources distributed
on the territory with a hierarchical organization, few qualified human resources)
1
.
We believe that a typical solution to perform business intelligence analysis realized
for use in developed countries cannot be applied as-is in developing countries because
the problems, the needs and the constraints are completely different. For that reason,
we choose to work in strict collaboration with local companies and governmental
organizations to bring our experience in the development of similar solutions in such
critical context: which only domain experts living directly in contact with these
problems could help us to fully understand.
1
Our solution brings new opportunities for the use of the same platform also in other domains,
such as economy and education.
2 Related Work
Mature technologies and methodologies are available in literature [2][13] for the
design and development of data warehouse (DWH) systems that may serve our data
analysis problem. However, these approaches make some assumptions that are not
valid in our context. First of all, they assume data can be extracted from IT systems at
the data sources and centralized in a data warehouse. The problem is that in our case
such IT systems are not always available (as in most of the cases they are not even
present) and there is not (yet) a central data collector that could host a DWH.
Furthermore, classical DWH approaches assume an enterprise is highly motivated
to create and maintain active a DWH sending timely information. In our case people
acting as data collectors may be: not only employees of an organization but also
volunteers, that should be motivated to collect timely and correct information; even if
they do not understand or see immediately the positive effect of their work.
Another limitation is that these solutions typically assume a DWH is loaded basically
in a real time mode, but in our case data may arrive with delays of weeks or not arrive
at all.
Solutions for Business Intelligence available in the open source community like
SpagoBI
2
or Talend
3
are interesting but too complex for our case as they provide
many functionalities that in these contexts are not useful, and they risk to transform
the solution into an unmanageable tool that nobody is able to use. Instead, we need to
aim at simplicity with few and intuitive functionalities. Furthermore, these systems
are general purpose BI applications that cannot be easily customized and cannot be
adapted easily to different contexts. Besides, they require an amount of resources that
in our case are not available; such as memory occupancy, local database and trained
people to configure and customize the Extract, Transform and Load (ETL) [14] and
reporting phases.
Developing an information integration solution using poor data quality is totally
useless and also risky as it may lead to wrong results from the analysis and
consequently wrong decisions. Conscious of this problem, we are currently studying
approaches to trace and improve data quality and in particular to: keep the consumers
of BI analysis (BI consumers) informed on the quality of the results they are using;
trace and identify the sources and causes of mistakes with the help of the BI
consumers by colleting suggestions on how to correct the mistakes; minimize the
occurrence of the mistakes with tools to support critical phases of the data lifecycle,
like data entry which is a task essentially with manual effort which is the primary
cause of inconsistent data.
In this regard, we will pay a particular attention to the data entry and the cleaning
phases and to the solutions already proposed in the research arena in the hope to reuse
and extend them. For example, in [11] is proposed Usher, an end-to-end system to
design forms that dynamically adapt to reduce the probability of errors during data
entry (e.g. providing feedback in real-time to guide the data enterer toward more
likely values).
2
http://www.spagoworld.org/xwiki/bin/view/SpagoBI/
3
http://www.talend.com/
The work in [1] extends the traditional Functional Dependencies into Conditional
functional dependencies (CFDs) to capture the notion of “correct data” and to
improve Data Cleaning tools in the detection and correction of inconsistencies and
errors in the data.
As pointed out in [3] dirty data
4
often violates some integrity constraints reflecting
organization’s policies related to the quality of the data. To deal with this, [3] presents
an approach that suggests possible rules and identifies conformant or non-conformant
records (context-dependent rules).
3 Solution
Currently, healthcare services are monitored by collecting (often manually) data on
the clinical process and sending hardcopy reports to districts, provinces and finally to
the healthcare office (MISAU). Owing to some lacks on IT-systems and connections,
data capture and analysis operations cannot always guarantee reliable and coherent
information.
Ministry of Health
SIS-H
Province
SIS-H
District
SIS-H
SIS-H Application
Hospital
C
o
n
t
a
i
n
e
r
Persistence
Manager
ebXML
Repository
Data
Entry
Import
(ETL)
Export BI
P
l
u
g
i
n
s
SIS-H
Fig. 1. Deployment of the SIS-H module in the healthcare hierarchical organization of
Mozambique.
We provide a generic communication infrastructure, SIS-H, to exchange data
amongst facilities, across the whole healthcare system hierarchy – from operational
units, hospitals, districts, provinces and up to the ministry – regardless of the
available communication media (e.g., compact disk, usb stick, web-internet).
The SIS-H module could be replicated at the different sites of the Mozambique
Healthcare organization and be applied also in other domains (e.g. to support the
4
Not suitable for data analysis, with data problems such as inconsistency and incompleteness.
Ministry of Education and of the Interior). The caption in (Fig. 1) shows the main
components of the module divided into a container managing data persistence
(ebXML Repository) and a series of plugins providing the applicative functionalities
and in particular: Data Entry, to support users to type in the system aggregated
statistics on patient admissions and discharges that are collected on paper forms;
Import (ETL), to load statistic data collected from other sites, typically from other
levels of the organization; Export, to prepare an export of statistical information to be
imported and used in another site, typically at a higher level of the hierarchy; BI, a
module to produce business intelligence reports.
The next section presents a more detailed description of how the architecture works
and in particular how the SIS-H module can produce an export that is understandable
to installations in other sites allowing their interoperability.
3.1 Architecture
Each SIS-H plugin serves a dedicated purpose independently from the rest of the
components so that modules can be changed or replaced without any impact on the
rest of the system. This ensures greater flexibility, scalability and maintainability of
the system. The main components of the architecture are the container and the various
plugins for data Entry, data import (ETL), data export, and BI purposes.
The core functionalities of the system are given by the container which
encapsulates a persistence manager (see Fig. 1). The Container is decomposed into
five pillar features namely: Plugin Manager, Workflow Manager, Data
mapping/conversion Manager and Plugin Installation Manager.
The Plugin Manager is responsible for associating user actions to dedicated
plugins as well as loading configurations of the selected plugin. The Workflow
Manager executes the workflow of the plugin by calling exposed methods through
java reflection. The Data mapping/conversion manager transforms plugin internal
data objects into the ebXML format (defined in [8]). Finally, the Plugin installation
manager deals with addition and removal of plugins into the system. The behaviour of
a plugin in terms of functionalities accessible by a certain user, data accessible and
output formats are configured in the Plugin installation manager and can be easily
updated to adapt to different contexts.
The SIS-H modules could be deployed in three flavours: centralized, client-server
and embedded. In the centralized configuration, the basic layer contains the ebXML
repository for data persistency while in the distributed case the ebXML repository is
hosted in a remote application server as in Fig. 2. Finally, the embedded configuration
is released with an embedded database (like Apache Derby
5
) for a fully portable
system. The system can work both in a loosely connected environment and in a fully
networked area
6
.
5
http://db.apache.org/derby/
6
On one hand, in remote areas where the network connectivity may be poor or even not be
present, a fat client approach in which the application and persistency layer reside on the
local machine or are embedded in the application would be suitable. On the other hand, in
larger cities such as Maputo (the capital of Mozambique), equipped with network
connectivity, the system can run in a client-server mode with LAN or internet configuration.
Fig. 2. Deployment diagram of the SIS-H application.
It is worth mentioning that the application can support different languages and
geographic regions through the use of ResourceBundle [12] residing outside the
architecture, thus new languages can be added without changing the overall
architecture of the system.
3.2 Conceptual Model
Fig. 3 shows the organizational model of the hierarchical structure of the Mozambican
health system. The healthcare facilities are organized in Health Care Units (health
centers and clinics) and hospitals of different kinds (Specialized Hospital, Rural
Hospital and so on). Each hospital is divided into departments which offer specific
services. Each healthcare facility refers to an entity of the Government (e.g. the
Province or MISAU) to which it must submit reports (known as “ficha”) on the
activities performed. The data structure of such reports reflects at a conceptual level
the internal operating procedures of hospitals and it is mapped internally and
transferred from one level of the organizational hierarchy to the next in XML
according to the ebXML standard.
Since there is no centralized database in which all the data is stored, it is vital not
only to identify each ficha uniquely across different health units all the way up to the
government administrative domain, but also to capture the organizational structure
and the hierarchical level from which the ficha originates. In essence, each SIS-H
module is aware of its location in the hierarchy of the health system. In this way it is
simple to map other hierarchical organizations (for example, to add another
hierarchical level, like an internal department or another top level administrative body
like another ministry).
The adopted encoding for uniquely indentifying the ficha follows the hierarchy of
the structure up to the service point in which the data was transferred. For example, if
the operator that performs data export is registered in Maputo under Medicine
department in section Surgery the unique identity of the ficha transferred would be as
shown by the path element in Fig. 4.
Fig. 3. Health system organizational model and relationships between units and structures of
government.
Fig. 4. Ficha flow and embedded organizational structure unique identity.
4 Deployed Prototype
All the system is designed and developed according to open source principles and
technologies. Fig. 5 shows the main form of the application to access to the following
features: the Data Entry plugin, used for the daily registration of admissions and
discharges of patients; the Export Plugin used to export the “fichas” XML files to
exchange data; the Import Plugin to import the data previously exported by a different
location; the BI Plugin, used to produce BI reports, in which we adopted Jasper
Intelligence
7
as BI tool; and System Management used to manage the users, the
general system configuration, backup procedures, etc.
Fig. 5. Main Form sample, and the access to the Plugins.
5 Conclusion and future work
Our modular architecture divided into plugin modules allows composing a suite of
functionalities on the base of the needs and characteristic of the installation site.
Plugins can be easily updated and configured to customize their behaviour and
functionalities, data accessible and output formats to adapt to different contexts. The
modular architecture allows for simple integration of specific functions based on flow
type, control functions and validation information.
In the future, our architecture could be extended to manage other protocols and
media, enabling the architecture to be applied in other contexts like the ministries of
7http://www.jaspersoft.com/
Export Plugin Import Plugin
System
Management
Data Entry Plugin
BI Plugin
finance and education. Other extensions could introduce the research results on the
quality of the data managed inside our architecture.
Acknowledgments
The project is headed by the Mozambican Ministry of Health (MISAU), collaborating
with the Department for Information and Health (DIS), MOASIS (closely related with
the University E. Mondlane) and the Consortium Pandora Box Ltda – GPI Spa. Our
special thanks to Dra. Célia Gonçalves, Dr. Amisse Momade, all team at MISAU,
Dr. Alessandro Campione and all MOASIS team.
References
1. P. Bohannon, W. Fan, F. Geerts, X. Jia, and A. Kementsietsidis, “Conditional functional
dependencies for data cleaning,” Proc. ICDE, 2007.
2. The Data Warehouse Toolkit: The Complete Guide to Dimensional Modeling, 2nd Edition.
John Wiley & Sons, 2002
3. Fei Chiang , Renée J. Miller, “Discovering data quality rules”, Proceedings of the VLDB
Endowment, 2008.
4. World Health Organization. ICD-10: International Statistical Classification of Diseases and
Related, Health Problems. 10th Revision.http://apps.who.int/classifications/apps/icd/icd10online/
5. World Health Organization. Mozambique, Mozambique's health system: Health and
development.http://www.who.int/countries/moz/areas/health_system/en/index1.html
6. Ministério da Saúde, Plano Estratégico do sector Saúde, 2007 -2012, República de
Moçambique.http://www.who.int/countries/moz/publications/pess_2007_2012.pdf
7. Ministry of health. Report on the revision of the data from HIV epidemiological
surveillance. Round 2007. February 2008. Republic of Mozambique.http://www.misau.gov.mz/pt/content/download/4637/28101/file/Ronda2007EN.pdf
8. Standard ebXML: Registry Information Model.http://www.ebxml.org/specs/index.htm#technical_specifications
9. World Health Organization. Terms of Reference for Designing the Requirements of the
Health Information System of the Maputo Central Hospital and preparation of the Tender
Specifications. Technical Report. Version 2.0. January 2007
10. Use of ICD-10 for morbidity and mortality notification for in-patients, in recourse limited
settings. The experience of Mozambique using reduced disease lists. Roberta Pastore,
Alessandro Campione, Bernardina Gonçalves, Armando Melo, Celia Goncalves, Carla Silva
Matos, Marcelino Mugai. Annual meeting of the WHO Family of International
Classifications Network
11. Chen K, Chen H, Conway N, Hellerstein J. Usher: Improving data quality with dynamic
forms. ICDE. 2010.
12. Sun Microsystems. Java Resource Bundle. [Online]http://java.sun.com/j2se/1.4.2/docs/api/java/util/ResourceBundle.html
13. Chaudhuri S, Dayal U. An overview of data warehousing and OLAP technology. ACM
Sigmod record. 1997.
14. R. Kimball and J. Caserta, The Data Warehouse ETL Toolkit: Practical Techniques for
Extracting, Cleaning, Conforming, and Delivering Data. John Wiley & Sons, 2004.
doc_942359834.pdf
Enabling Business Intelligence Functions over a Loosely Coupled Environment
Enabling Business Intelligence Functions over a Loosely
Coupled Environment
Giampaolo Armellin
1
, Leandro Paulo Bogoni
1,2
, Annamaria Chiasera
1,2
,
Tefo James Toai
1
, Gianpaolo Zanella
1
1
GPI SpA, Via Ragazzi del '99, 13 - Trento, Italy
2
Information Engineering and Computer Science Department - University of Trento,
Trento, Italy
[email protected],
{garmellin, achiasera, ttoai, gzanella}@gpi.it
Abstract. Planning effective and well targeted actions to manage and improve
the local and national healthcare services requires institutions to understand and
analyse the real needs of the population based on reliable and timely statistical
analysis on citizens’ health state. This is particularly important in developing
countries in which healthcare facilities lack ICT infrastructures and network
connectivity, making data collection and analysis particularly difficult with a
considerable manual effort leading to potentially unreliable or incoherent
information. In this scenario, we propose a generic communication
infrastructure, developed in the SIS-H project for Mozambique hospitals to
capture, communicate and analyse clinical events. Our solution enables the
exchange of data amongst healthcare facilities over all the different aggregation
levels of the hierarchical healthcare system of Mozambique regardless of the
availability of communication media (e.g., compact disk, usb stick, web-
internet). The plugin-based solution adopted supports reporting and Business
Intelligence analysis for exploring data at different granularity levels.
Keywords: Business Intelligence, e-Governance, plugin-based architecture
1 Introduction
Improving accessibility and quality of health services is one of the outcomes of the
Mozambican Health Sector Strategic Plan (PESS 2007-2012 6, 5). Despite public and
external founds, medical assistance is not accessible to all the population owing to the
difficulty in coordinating and controlling the actions performed and a chronicle
deficiency of human resources. The World Health Organization points out the need of
monitoring systems to control resource flows, progresses and outcomes finalized to
monitor and evaluate health services [5], [9]. In the Strategic Plan for the Health
Sector [6] the role of such monitoring system is essential for decision makers to
manage and supervise the maintenance of hospitals. Another pre-requisite to evaluate
and improve health quality is to encourage community participation with the active
communication of the results of the analysis promoting transparency and
accountability.
In this paper we will report our experience in the development of a communication
infrastructure applicable at the national level for the monitoring of the quality of
healthcare system in Mozambique. We propose a lightweight and robust infrastructure
that requires minimum resources to operate with poor or no connectivity to support
people in the collection, processing and communication of statistical information with
a simple and intuitive tool usable even by not trained personnel.
1.1 The Context
The structure of the Mozambican Healthcare System is level-oriented and hierarchical
– i.e. operational units and hospitals are grouped according to managed services and
geographical areas. The central hospitals of Beira and Maputo are at the top levels of
the hierarchy (level IV hospitals) and refer directly to the ministry of health. Pemba
provincial hospital is at level III and refers to the Provincial Health Office.
The ministry of health and the provinces provide economical support to operational
units and hospitals and plan actions and projects to prevent and deal with chronicle
and highly contagious diseases (like Malaria, HIV/AIDS, Cholera). These activities
include, among others, economical investments to improve the infrastructures
(hospitals’ facilities like emergency rooms, intensive units, infective divisions), staff
training, and education of the population for prevention.
Governing bodies need to know which are the real needs of the population to
identify and deal with sanitary emergency, to plan an effective set of activities (e.g.
the continuous and periodic surveillance required by HIV epidemic [7]) and to reduce
waste of resources (staff and financial). In [10] it is highlighted how this requires at
first the surveillance of the morbidity and mortality causes. The analysis of such
statistical data allows on one side to understand the health state of the population and
on the other side to monitor hospital resource utilization and in particular costs of
hospitalization and staff employed. Based on these analyses the ministry of health can
better plan public health interventions, for example to grant a suitable number of beds
in hospitals or health posts facing particularly critical sanitary crisis.
As pointed out in the analysis of Campione et al. [10] the social, economical and
technological environment of Mozambique poses some strict constraints to realize
such analysis that in developed countries are normal Business Intelligence activities
[2]. In particular [10] identifies the following challenges:
? lack of IT skills: qualified personnel, able to operate IT devices and tools is scarce
and consequently the data collected is not trustworthy;
? lack of data: the diagnostic capability of the lower level point of care is rather poor
compared to more specialized level of care (hospitals of level III or below have
less diagnostic facilities than hospitals of level IV);
? territory highly distributed and sparsely populate: in order to get a comprehensive
picture of the health state of population it is necessary to have a good coverage of
the territory at the national level;
? infrastructure limitations: no electricity, no information communication
technologies, no connectivity and no computers available;
? lack of information systems: there are neither the applications to manage health
records nor the personal information of patients and consequently it is difficult to
identify patients in the long term and to maintain detailed information on their
health story.
Data is typically collected on paper with significant manual effort and difficulties
of communication and analysis by other parties at higher level of the hierarchy where
planning and resource management is performed. The lack of computerization in the
point of care, especially at the periphery, makes the data collection particularly
difficult; with a considerable amount of time and manual effort required from the
personnel. Furthermore, people are not motivated to spend effort in such a boring task
as they do not see any useful results in short time coming out. These factors leave the
data collection process incomplete and error-prone.
Indeed a great step forward has been performed with the adoption of a standardized
list of diagnosis selected from the ICD-10 standard [4]. However, an IT solution
capable of effectively supporting the collection of data, and is easy to install in
critical environment with lack of infrastructure and is simple to use even for poorly
trained people; is still missing. This solution should be general enough to be usable at
any level of the national health care system with a great degree of flexibility and
customizability to collect as much information as possible.
In this work we present the solution we provide in the context of the project
“SIS-H – Módulo Internamentos” launched with internal tender by the Mozambican
Ministry of Health (MISAU) according to the Mozambique eGovernment Strategy.
The main project goal is to rapidly overcome the lack of statistical information for
hospitals of level III and level IV, devising software applications to collect and
aggregate data in the clinical area, with specific focus on admissions and discharges
of patients. Those aggregated data concern Maputo Central Hospital, Beira Central
Hospital and Pemba Provincial Hospital and are being periodically sent to Provincial
Health Offices and to the Ministry of Health, enabling analysis and planning on the
health care system.
We provide a platform that effectively answers the challenges identified above
both technical (lack of IT infrastructure) and organizational (data sources distributed
on the territory with a hierarchical organization, few qualified human resources)
1
.
We believe that a typical solution to perform business intelligence analysis realized
for use in developed countries cannot be applied as-is in developing countries because
the problems, the needs and the constraints are completely different. For that reason,
we choose to work in strict collaboration with local companies and governmental
organizations to bring our experience in the development of similar solutions in such
critical context: which only domain experts living directly in contact with these
problems could help us to fully understand.
1
Our solution brings new opportunities for the use of the same platform also in other domains,
such as economy and education.
2 Related Work
Mature technologies and methodologies are available in literature [2][13] for the
design and development of data warehouse (DWH) systems that may serve our data
analysis problem. However, these approaches make some assumptions that are not
valid in our context. First of all, they assume data can be extracted from IT systems at
the data sources and centralized in a data warehouse. The problem is that in our case
such IT systems are not always available (as in most of the cases they are not even
present) and there is not (yet) a central data collector that could host a DWH.
Furthermore, classical DWH approaches assume an enterprise is highly motivated
to create and maintain active a DWH sending timely information. In our case people
acting as data collectors may be: not only employees of an organization but also
volunteers, that should be motivated to collect timely and correct information; even if
they do not understand or see immediately the positive effect of their work.
Another limitation is that these solutions typically assume a DWH is loaded basically
in a real time mode, but in our case data may arrive with delays of weeks or not arrive
at all.
Solutions for Business Intelligence available in the open source community like
SpagoBI
2
or Talend
3
are interesting but too complex for our case as they provide
many functionalities that in these contexts are not useful, and they risk to transform
the solution into an unmanageable tool that nobody is able to use. Instead, we need to
aim at simplicity with few and intuitive functionalities. Furthermore, these systems
are general purpose BI applications that cannot be easily customized and cannot be
adapted easily to different contexts. Besides, they require an amount of resources that
in our case are not available; such as memory occupancy, local database and trained
people to configure and customize the Extract, Transform and Load (ETL) [14] and
reporting phases.
Developing an information integration solution using poor data quality is totally
useless and also risky as it may lead to wrong results from the analysis and
consequently wrong decisions. Conscious of this problem, we are currently studying
approaches to trace and improve data quality and in particular to: keep the consumers
of BI analysis (BI consumers) informed on the quality of the results they are using;
trace and identify the sources and causes of mistakes with the help of the BI
consumers by colleting suggestions on how to correct the mistakes; minimize the
occurrence of the mistakes with tools to support critical phases of the data lifecycle,
like data entry which is a task essentially with manual effort which is the primary
cause of inconsistent data.
In this regard, we will pay a particular attention to the data entry and the cleaning
phases and to the solutions already proposed in the research arena in the hope to reuse
and extend them. For example, in [11] is proposed Usher, an end-to-end system to
design forms that dynamically adapt to reduce the probability of errors during data
entry (e.g. providing feedback in real-time to guide the data enterer toward more
likely values).
2
http://www.spagoworld.org/xwiki/bin/view/SpagoBI/
3
http://www.talend.com/
The work in [1] extends the traditional Functional Dependencies into Conditional
functional dependencies (CFDs) to capture the notion of “correct data” and to
improve Data Cleaning tools in the detection and correction of inconsistencies and
errors in the data.
As pointed out in [3] dirty data
4
often violates some integrity constraints reflecting
organization’s policies related to the quality of the data. To deal with this, [3] presents
an approach that suggests possible rules and identifies conformant or non-conformant
records (context-dependent rules).
3 Solution
Currently, healthcare services are monitored by collecting (often manually) data on
the clinical process and sending hardcopy reports to districts, provinces and finally to
the healthcare office (MISAU). Owing to some lacks on IT-systems and connections,
data capture and analysis operations cannot always guarantee reliable and coherent
information.
Ministry of Health
SIS-H
Province
SIS-H
District
SIS-H
SIS-H Application
Hospital
C
o
n
t
a
i
n
e
r
Persistence
Manager
ebXML
Repository
Data
Entry
Import
(ETL)
Export BI
P
l
u
g
i
n
s
SIS-H
Fig. 1. Deployment of the SIS-H module in the healthcare hierarchical organization of
Mozambique.
We provide a generic communication infrastructure, SIS-H, to exchange data
amongst facilities, across the whole healthcare system hierarchy – from operational
units, hospitals, districts, provinces and up to the ministry – regardless of the
available communication media (e.g., compact disk, usb stick, web-internet).
The SIS-H module could be replicated at the different sites of the Mozambique
Healthcare organization and be applied also in other domains (e.g. to support the
4
Not suitable for data analysis, with data problems such as inconsistency and incompleteness.
Ministry of Education and of the Interior). The caption in (Fig. 1) shows the main
components of the module divided into a container managing data persistence
(ebXML Repository) and a series of plugins providing the applicative functionalities
and in particular: Data Entry, to support users to type in the system aggregated
statistics on patient admissions and discharges that are collected on paper forms;
Import (ETL), to load statistic data collected from other sites, typically from other
levels of the organization; Export, to prepare an export of statistical information to be
imported and used in another site, typically at a higher level of the hierarchy; BI, a
module to produce business intelligence reports.
The next section presents a more detailed description of how the architecture works
and in particular how the SIS-H module can produce an export that is understandable
to installations in other sites allowing their interoperability.
3.1 Architecture
Each SIS-H plugin serves a dedicated purpose independently from the rest of the
components so that modules can be changed or replaced without any impact on the
rest of the system. This ensures greater flexibility, scalability and maintainability of
the system. The main components of the architecture are the container and the various
plugins for data Entry, data import (ETL), data export, and BI purposes.
The core functionalities of the system are given by the container which
encapsulates a persistence manager (see Fig. 1). The Container is decomposed into
five pillar features namely: Plugin Manager, Workflow Manager, Data
mapping/conversion Manager and Plugin Installation Manager.
The Plugin Manager is responsible for associating user actions to dedicated
plugins as well as loading configurations of the selected plugin. The Workflow
Manager executes the workflow of the plugin by calling exposed methods through
java reflection. The Data mapping/conversion manager transforms plugin internal
data objects into the ebXML format (defined in [8]). Finally, the Plugin installation
manager deals with addition and removal of plugins into the system. The behaviour of
a plugin in terms of functionalities accessible by a certain user, data accessible and
output formats are configured in the Plugin installation manager and can be easily
updated to adapt to different contexts.
The SIS-H modules could be deployed in three flavours: centralized, client-server
and embedded. In the centralized configuration, the basic layer contains the ebXML
repository for data persistency while in the distributed case the ebXML repository is
hosted in a remote application server as in Fig. 2. Finally, the embedded configuration
is released with an embedded database (like Apache Derby
5
) for a fully portable
system. The system can work both in a loosely connected environment and in a fully
networked area
6
.
5
http://db.apache.org/derby/
6
On one hand, in remote areas where the network connectivity may be poor or even not be
present, a fat client approach in which the application and persistency layer reside on the
local machine or are embedded in the application would be suitable. On the other hand, in
larger cities such as Maputo (the capital of Mozambique), equipped with network
connectivity, the system can run in a client-server mode with LAN or internet configuration.
Fig. 2. Deployment diagram of the SIS-H application.
It is worth mentioning that the application can support different languages and
geographic regions through the use of ResourceBundle [12] residing outside the
architecture, thus new languages can be added without changing the overall
architecture of the system.
3.2 Conceptual Model
Fig. 3 shows the organizational model of the hierarchical structure of the Mozambican
health system. The healthcare facilities are organized in Health Care Units (health
centers and clinics) and hospitals of different kinds (Specialized Hospital, Rural
Hospital and so on). Each hospital is divided into departments which offer specific
services. Each healthcare facility refers to an entity of the Government (e.g. the
Province or MISAU) to which it must submit reports (known as “ficha”) on the
activities performed. The data structure of such reports reflects at a conceptual level
the internal operating procedures of hospitals and it is mapped internally and
transferred from one level of the organizational hierarchy to the next in XML
according to the ebXML standard.
Since there is no centralized database in which all the data is stored, it is vital not
only to identify each ficha uniquely across different health units all the way up to the
government administrative domain, but also to capture the organizational structure
and the hierarchical level from which the ficha originates. In essence, each SIS-H
module is aware of its location in the hierarchy of the health system. In this way it is
simple to map other hierarchical organizations (for example, to add another
hierarchical level, like an internal department or another top level administrative body
like another ministry).
The adopted encoding for uniquely indentifying the ficha follows the hierarchy of
the structure up to the service point in which the data was transferred. For example, if
the operator that performs data export is registered in Maputo under Medicine
department in section Surgery the unique identity of the ficha transferred would be as
shown by the path element in Fig. 4.
Fig. 3. Health system organizational model and relationships between units and structures of
government.
Fig. 4. Ficha flow and embedded organizational structure unique identity.
4 Deployed Prototype
All the system is designed and developed according to open source principles and
technologies. Fig. 5 shows the main form of the application to access to the following
features: the Data Entry plugin, used for the daily registration of admissions and
discharges of patients; the Export Plugin used to export the “fichas” XML files to
exchange data; the Import Plugin to import the data previously exported by a different
location; the BI Plugin, used to produce BI reports, in which we adopted Jasper
Intelligence
7
as BI tool; and System Management used to manage the users, the
general system configuration, backup procedures, etc.
Fig. 5. Main Form sample, and the access to the Plugins.
5 Conclusion and future work
Our modular architecture divided into plugin modules allows composing a suite of
functionalities on the base of the needs and characteristic of the installation site.
Plugins can be easily updated and configured to customize their behaviour and
functionalities, data accessible and output formats to adapt to different contexts. The
modular architecture allows for simple integration of specific functions based on flow
type, control functions and validation information.
In the future, our architecture could be extended to manage other protocols and
media, enabling the architecture to be applied in other contexts like the ministries of
7http://www.jaspersoft.com/
Export Plugin Import Plugin
System
Management
Data Entry Plugin
BI Plugin
finance and education. Other extensions could introduce the research results on the
quality of the data managed inside our architecture.
Acknowledgments
The project is headed by the Mozambican Ministry of Health (MISAU), collaborating
with the Department for Information and Health (DIS), MOASIS (closely related with
the University E. Mondlane) and the Consortium Pandora Box Ltda – GPI Spa. Our
special thanks to Dra. Célia Gonçalves, Dr. Amisse Momade, all team at MISAU,
Dr. Alessandro Campione and all MOASIS team.
References
1. P. Bohannon, W. Fan, F. Geerts, X. Jia, and A. Kementsietsidis, “Conditional functional
dependencies for data cleaning,” Proc. ICDE, 2007.
2. The Data Warehouse Toolkit: The Complete Guide to Dimensional Modeling, 2nd Edition.
John Wiley & Sons, 2002
3. Fei Chiang , Renée J. Miller, “Discovering data quality rules”, Proceedings of the VLDB
Endowment, 2008.
4. World Health Organization. ICD-10: International Statistical Classification of Diseases and
Related, Health Problems. 10th Revision.http://apps.who.int/classifications/apps/icd/icd10online/
5. World Health Organization. Mozambique, Mozambique's health system: Health and
development.http://www.who.int/countries/moz/areas/health_system/en/index1.html
6. Ministério da Saúde, Plano Estratégico do sector Saúde, 2007 -2012, República de
Moçambique.http://www.who.int/countries/moz/publications/pess_2007_2012.pdf
7. Ministry of health. Report on the revision of the data from HIV epidemiological
surveillance. Round 2007. February 2008. Republic of Mozambique.http://www.misau.gov.mz/pt/content/download/4637/28101/file/Ronda2007EN.pdf
8. Standard ebXML: Registry Information Model.http://www.ebxml.org/specs/index.htm#technical_specifications
9. World Health Organization. Terms of Reference for Designing the Requirements of the
Health Information System of the Maputo Central Hospital and preparation of the Tender
Specifications. Technical Report. Version 2.0. January 2007
10. Use of ICD-10 for morbidity and mortality notification for in-patients, in recourse limited
settings. The experience of Mozambique using reduced disease lists. Roberta Pastore,
Alessandro Campione, Bernardina Gonçalves, Armando Melo, Celia Goncalves, Carla Silva
Matos, Marcelino Mugai. Annual meeting of the WHO Family of International
Classifications Network
11. Chen K, Chen H, Conway N, Hellerstein J. Usher: Improving data quality with dynamic
forms. ICDE. 2010.
12. Sun Microsystems. Java Resource Bundle. [Online]http://java.sun.com/j2se/1.4.2/docs/api/java/util/ResourceBundle.html
13. Chaudhuri S, Dayal U. An overview of data warehousing and OLAP technology. ACM
Sigmod record. 1997.
14. R. Kimball and J. Caserta, The Data Warehouse ETL Toolkit: Practical Techniques for
Extracting, Cleaning, Conforming, and Delivering Data. John Wiley & Sons, 2004.
doc_942359834.pdf