Supporting ambulant care through information technologies will assist the care personnel in their everyday routine resulting in more time for their patients. Therefore the MARIKA project focusses on developing an assistance system for facilitating the care documentation process through pervasive technologies. In this context, various data need to be handled which requires database techniques for di erent aspects of the assistance system. Thus a general ve layer architecture for assistance systems is introduced and adapted according to the requirements of data management. With the focus on the layer of Mobile Content Management it is explained how a part of the distribution of the data in the MARIKA project is handled.
distributed database design
Recent statistical studies in the USA as well as in European
countries estimate a dramatic increase in the group of the
elderly in the next decades [
With respect to assistive care systems, low level data which are only stored temporarily for e ciency as well as high level data for the actual care documentation need to be handled. Additionally already known information can contribute to the analysis of the situation for the assistance system. To manage all these data, di erent database techniques are applicable. After a short introduction on assistance systems for home care support, this paper proposes database techniques for di erent purposes in an assistance system on the basis of a general ve layer architecture for assistance systems. Focussing on the requirement to o er the correct information needed at the right time in the right place, the subsequent section deals with the distribution of data in the MARIKA scenario. After that the conclusion is presented. Finally an outlook will be given in the last section.
Considering the demographic change, support for the elderly
and those who care for them is inevitable. Assistance
systems can help the elderly to live longer at home
independently. Various research projects deal with environments to
o er assistive technologies, e.g. via a mobile device which
needs to be carried [
In the area of home care support the MARIKA project deals
with the development of an assistance system for
supporting home care personnel [
Sensing and Positioning
information about the patients and the planned care
activities. The performed care activities are recognized with the
help of sensors and matched to the planned activities.
Other projects also aim to support the care personnel. One
example is the SAMDY project which intends to establish an
early warning system for the care personnel by monitoring
the elderly through sensors. It also supports care
documentation with sensors but is intended to be used for stationary
care mainly [
To emphasize the charectaristica of a general assistance
system, a general architecture for assistance systems as shown
in gure 1 has been proposed in [
The rst layer which delivers the basic data for analysing
the user's situation and acitvities is named Sensing and
Positioning. Sensors (e.g. acceleration sensors, RFID tags)
and positioning systems (e.g. GPS for outdoors) collect raw
sensor data which are parsed and prepared for the situation
and activity analysis on a higher level. For supporting these
processes a tiny database system may be used to manage
the sensor data, like the smallest version possible of
FAMEDBMS [
The Heterogeneous Networking and Ad-hoc Device Cooperation layer is responsible for integrating the heterogeneous devices and connect the necessary subsystems. The devices may not only di er in their type, but also in their communication technology and their protocol speci cations. Additionally it is possible that the needed and the available devices are not known in advance. This adhoc device cooperation can be realized with a service-oriented approach (e.g. SOA). One idea to solve the heterogeneity in the communication technology of a service consumer and a service Multimodal Interaction Users - Inputs/Outputs Multimodale Interaktion
- data from keyboards, microphones, cameras, ...
Situations- und HandlungSsvituoarthioenrsaangdeActivity Prediction
Situation analysis - domain knowledge/ontologies Mobiles Content-Management
Activity analysis - high level models - activity diagrams Heterogene VernetzMuonbgile&Content Management Prediction Spontane Gerätekooperation
Sensorik und Ortung
Data Integration
Home InformationBase
Replication+Synchronisation
Database Server Heterogeneous Networking and Ad-Hoc Device Cooperation
Replication+Synchronisation
GPAP SOA Communication - Bluetooth-, Webservices Sensing and Positioning
Communication sensor-data sesnesnosro-rd-adtaata
Tiny DB
GPS
raw stream data management
- parsing, aggregation, ...
provider is to introduce a tier of abstraction. It converts the
used precise technology into an abstract one and transforms
this into all other present technologies. One example is the
General Purpose Access Point (GPAP) which is developed
by Dressler et.al.[
The main purpose of the Mobile Content Management is to provide the correct information needed in the right place at the right time. This requires a prediction about which information may be needed at which place in the future. Therefore the analysis and prediction of situations and activities, which is done in this layer, is highly important. Furthermore the Mobile Content Management comprises the management of all multimedia documents and the replication of those to ful ll the main purpose. Therefore a medium to very large2 database system, which will probably be distributed or federated to ensure the information to be in the right place when needed, is recommended. The integration of necessary other information, database or assistance systems is another task of this layer.
The Situation and Activity Prediction layer is the key layer of an assistance system. Only if the situation or activitiy of the user is understood and the possible future situation or activity is recognized then the system can o er assistance. 2depending on the amount of multimedia documents So, this layer is responsible for analysing the user's situation or activity with the help of the data from the Sensing and Positioning layer and calculates probabilities for the future situations or activities. It utilizes high level models and/or ontologies. Surrounding conditions and prior knowledge additionally help identify the situation. Because the situation analysis runs permanently the necessary models and ontologies need to reside in the memory all the time. Therefore a database system at this point is only helpful for storing some background information which is not constantly needed and can be queried if required.
The layer of Multimodal Interaction deals with the
communication of the assistance system with the user via
several input and output devices. The received input data via
di erent devices, e.g. keyboard, microphone, camera, may
in uence the situation analysis. A camera can for
example detect the movement of a person which helps identify
their situation. Concerning the input, a natural interaction
should be intended to avoid the feeling of awkwardness. If
gestures are necessary, they should be socially accepted ones,
like mentioned in [
Another architecture for assistance systems consisting of
only three tiers (Context Sensing Tier, Context
Processing Tier, Context Application Tier ) has been introduced by
Kurschl et al. [
While the architecture presented above can be applied for any kind of assistance system, this chapter focusses on one speci c layer in the scenario of home care support only: the Mobile Content Management in the MARIKA project. In this scenario the relevant content to be managed consists of the care documentation and all of its related data concerning care plan, care providers, patients, health and so on. These data comprise sensor data streams to recognize the activities, text and multimedia documents (e.g. pictures, videos, recorded speech) for complementing the documentation as well as temporal and spatial data. To manage these data a distributed database system is used which mainly consists of two parts: a central database which is installed for instance in the o ce of the care provider and several mobile databases which are located on the mobile devices of the care CareActivity CareActivity_Patient
Patient
Schedule CareActivity_Patient Nurse Timestamp
Nurse personnel. More databases may be included in this scenario, even a federated approach is possible if other independent database systems will be involved. But for the purpose of this paper the scenario with the above described two-sided distributed database system is su cient.
The care provider has created a care plan for each patient
that de nes which care activities have to be performed and
how often. Based on that a working schedule for all nurses
is developed which is stored in the central database. The
schedule can be seen as the set S fN; T; P; Cg where N is
the set of all nurses, T is the set of all timestamps, P is the
set of all patients and C is the set of all care activities. Each
tuple (n; t; p; c) 2 S with n 2 N , t 2 T , p 2 P and c 2 C
is unique. One nurse only needs a subset of this schedule
for one working shift to be replicated on their mobile
device. This subset R S is de ned by the speci c nurse
na 2 N as well as the start time tstart 2 T and end time
tend 2 T of their working shift for that day and contains
all tuples (n; t; p; c) which correspond to the following
condition: n = na and tstart t tend where n 2 N and t 2 T .
This subset and all corresponding data (like health
information about the client) need to be replicated to the database
on the mobile device. During their shift the nurse can
consult their mobile device for information about the patient
and the care activities. With the help of sensors the care
activities are recognised and the information on the mobile
device is updated. Eventually new health information about
the patient and some additional care information is added
by the nurse. Ideally, after each patient the updated mobile
database will be synchronised with the centralized database
in order to ensure that the information about the patient
is up-to-date for a possible other nurse who may be visiting
the patient a little later. Because the schedule itself contains
timestamps determining when each care activity is planned
to be performed with a patient, the synchronisations are
expected around a particular time. If one synchronisation is
missing, e.g. because of no network connectivity, the system
needs to handle that. A late synchronisation might lead to
con icts which need to be resolved. Classical replication
techniques like the Read-One-Write-All technique are not
suitable for a mobile replication. Du er et. al. [
The key characteristic of the database scheme is the realization of the schedule. Figure 3 shows a sketch of the database scheme involving ve relations (the whole database scheme contains more relations, for simplicity the gure only shows the ones needed to explain the implementation of the above described schedule). The care plan which describes which care activity is performed on which patient is contained in the relation CareActivity Patient. The schedule is represented by the relation Schedule and contains two foreign keys, one relating to the Nurse relation and one relating to the CareActivity Patient relation, as well as a timestamp. For replicating the relevant data to the nurses' mobile devices, a horizontal partitioning based on the speci c nurses and their working shifts as stated above serves as the main concept. Therefore every working day each nurse only carries the necessary information about their patients and the planned care activities.
All documentation about the performed care activities need to reside with the patient at least until the end of the month. Currently this documentation is paper-based. With the introduction of an assistive system to support home care like in the MARIKA project all information about the documentation is digitalised. Instead of printing it out and leaving the documentation on paper with the patient, a home information base can be installed in the patient's home to replace the paper work. The data on the device needs to be replicated from the central care information system or the mobile device (or both) and can be accessed via a display like the television. The relevant care information is considered as read-only. Therefore one question is whether it should be materialized on the home information base or whether it should only be integrated virtually.
Additionally, health data like vital data can be collected on the home information base. Furthermore, in the future it can also be used by other stakeholders like relatives or medical sta to store and access patient-related information. The home information base can be used as a integration platform for other possible services for the elderly. For integrating assistance systems the research area of information integration needs to be investigated to nd out whether and how these techniques can be utilized. Several special aspects have to be considered for assistance systems, e.g. how the functionality is integrated and whether the combination of di erent assistive functionality can cause some malfunctional or contradictory behavior. Another aspect is privacy of the patient's sensitive data. First of all the data itself needs to be encrypted. Also the access to the data needs to be restricted. Each stakeholder is only allowed to access a certain amount of data after authentication. So data security is one major issue when installing the storage device.
Home care support through assistance systems has been
elaborated as an important issue regarding the demographic
change. While various projects support the elderly directly
([
In the MARIKA project, activities of a home care nurse are detected in order to automate documentation of the home care process. To respect the privacy of the nurse, the sensor data has to be reduced as early as possible in the process of replication and distribution. If possible, the tiny DBMS managing the sensor data should perform as much aggregations and selections to only distribute those sensor data being important to detect a given home care activity. On the other hand, the tiny DBMS has not the power to derive the important home care activities: the data has to be distributed to other computers, e.g. the home information base. In the upcoming phase of the MARIKA project, we have to bridge the gap between good activity detection and preserving privacy of the home care personnel.
Another important future direction of research is the federation and evolution of assistance systems. The home information base mentioned above is primarily planned to be a basis for AAL (ambient assisted living) systems, i.e. it should support the people in their homes and not the nurses visiting their homes. If one computer system and one set of data is the basis for more than one assistance systems, the federation of the data has to be performed without having con icts with the privacy preservation for both user groups to be assisted. And if activities and home care standards change, the activity detection has to change and the aggregeation, selection and distribution of data has to automatically be adapted.