An Approach Towards Information Quality Management of Electronic

Health Records

Mohammad Mahdi Mahdavi Amjad, Kamyar Rasta, Martin Gerdes and Rune Fensli

Department of Information and Communication Technology, University of Agder, Grimstad, Norway

Keywords:

Electronic Health Record, Information Quality, EHR Analysis.

Abstract:

Electronic Health Records (EHR) management systems, have gained attention in many countries as public

organizations working in health domain (such as hospitals and municipalities) use them to lift the quality of

health care. EHRs are considered as the basis for establishing Health Information Service (HIS) systems. In

spite of remarkable advantages of EHRs, the lack of quality metrics can reduce the efﬁciency of high level

systems that are based on them. Particularly, it is smart to design a mechanism to assure the quality of data

generated by sensors. In this paper, we propose an architecture in which EHRs are enriched with metadata

to provide the information quality. Our architecture emphasizes on the quality of EHRs since we believe that

the quality aspect of health records has not been contemplated in many commercial systems. We utilize some

quality dimensions to produce the total quality metric. Moreover, we show that this architecture can provide

quality-based systems with more appropriate inputs.

1 INTRODUCTION

As computers are increasingly used in health care

realm, more applications are developed for managing

electronic health records (EHR). Utilizing the EHRs

have lifted the quality of health care (Gunter and

Terry, 2005)(DesRoches et al., 2008). There are nu-

merous applications out there and each one has its

own method to store and analyze EHRs. We have

studied some of them e.g. OpenMRS and OpenEHR

and noticed that in almost all of them there is no

mechanism for handling information quality. (Kalra

et al., 2005)(Seebregts et al., 2009) As EHRs are typ-

ically raw data of health sector analyzing systems, the

information quality of them is extremely important to

improve.

The quality of EHRs must be ensured before be-

ing used in other higher level systems. On the other

hand, the use of sensors has been drastically in-

creased in the e-Health area and it means that there

can be some errors in automatically generated mea-

surements. Hence, the necessity of normalizing the

automatically generated records shows itself.

In general, in case of using sensor networks, the

probability of receiving wrong data from the sub-

ject can dramatically increase, especially if the error

detection and error correction mechanisms have not

been implemented (Akyildiz et al., 2002). Addition-

ally, remarkable changes of the vital signs measure-

ments are not usual incidents and beside other param-

eters, can be signs of emergency condition. This sig-

niﬁcant feature, can be used in notiﬁcation systems

but currently implemented systems do not support

it, although there exist applications that have some

thresholds for data or triggered alarm systems.

In this paper, we present an architecture which en-

hances EHR management systems by mechanisms to

determine the information quality in terms of different

quality dimensions. This architecture can be used to

ensure that EHR data meet some basic criteria. More-

over, our system produces some metadata for each

EHR that helps analyzing systems to achieve better

performances.

2 INFORMATION QUALITY OF

EHRS

To the best of our knowledge, almost none of the ex-

isting EHR management systems have mechanisms to

determine the information quality of health records.

This can degrade the result of EHR analyses espe-

cially when machines are responsible for generating,

storing, and using them.

Majority of the EHR Management (EHRM) sys-

270

Mahdavi Amjad M., Rasta K., Gerdes M. and Fensli R..

An Approach Towards Information Quality Management of Electronic Health Records.

DOI: 10.5220/0004752402700275

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2014), pages 270-275

ISBN: 978-989-758-010-9

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

Figure 1: Conventional two-tier EHR management system.

tems use the conventional approach to manage the

medical records. As Figure 1 shows, typically there

is a relational database as a data storage space and a

web/form application as the user interface which en-

ables users to store and retrieve data into the storage

and creates reports.

We have developed an application which can nor-

malize EHRs in a relational database. This applica-

tion has a collection of rules that can be applied to al-

most all types of EHR. For instance, assume that we

have a system in which a sensor measures the body

temperature of a subject and sends it to the recording

system. The sensor stops functioning after a while

and generates fake data. This can severely affect the

quality of the health record and consequently the va-

lidity of analysis.

In spite of conventional approach, we propose

an architecture in which a Quality-Enhanced Broker

Server (Q-EBS) controls the EHR’s quality. The Q-

EBS is responsible for calculating and assigning the

quality metric of each EHR. The quality metric is a

value that represents the outcome of the quality di-

mension values. From information quality point of

view, there are several popular quality dimensions

(Wand and Wang, 1996) such as accuracy, timeliness,

completeness, consistency, access security, relevancy,

understandability, etc. which are used in many sys-

tems. However, in this paper we just deal with the

ﬁrst four and since connectivity is important in mobile

networks, we consider it as another quality dimension

as well. Other quality dimensions can be engaged if

necessary.

2.1 Quality Dimensions

The computation of the ﬁnal quality metric requires

the values of the quality dimensions. In this section

we brieﬂy deﬁne our quality dimensions and the cor-

responding calculation equations. There is no consen-

sus on the deﬁnitions of quality dimensions, and in

most cases the deﬁnitions are customized according

to the requirements of the system (Wand and Wang,

1996) (Baumgartner et al., 2010) (Strong et al., 1997).

2.1.1 Accuracy

The ﬁrst quality dimension that we propose is accu-

racy. In our system, accuracy shows the ratio of the

observed value and the real value. We name this real

value as reference value. In real world there is no way

to retrieve the reference (or the real) value. As any

other measurement, ours is just a reﬂection of truth

and therefore cannot be exactly true. One possible

way for generating the reference value in health do-

main is to ask a nurse to manually produce it. Hence,

we can consider this value as the reference and simply

ignore the measurement errors.

The reference value can be generated by the sys-

tem automatically as well. To do this, we should con-

sider a proper time for system to reach to a reference

value which can be simply the average of the values

received in a past period of time. Nevertheless, the

production process of a reliable reference value is out

of the scope of this paper and so we do not focus on

how to get the reference value. We just want to show

how to calculate the accuracy of EHRs based on a

given reference value.

Whether using an automatically or manually gen-

erated reference, we can compute the accuracy di-

mension values to ensure that incoming data is not

produced by a malfunctioning sensor. Eq. (1) is used

to calculate the accuracy of an observed value where

r represents the reference value and o represents the

observed value. Min and Max functions are utilized

for calculating the minimum and maximum values of

the reference and the observed values respectively.

Accuracy =

Min(r, o)

Max(r, o)

∗ 100 (1)

Note that in reality there is no practical way to

generate reference values for each EHR. Neverthe-

less, reference number is a concept that can be used

for evaluating the quality of an observed data.

2.1.2 Completeness

The second quality dimension we use is the complete-

ness. The completeness dimension, demonstrates the

proportion of the number of received data items and

the expected number of data items as a number be-

tween zero and hundred. In other words, the com-

pleteness shows how many data items are received

from the sensor and how many of them we expect to

receive. For example, assume that we have a device

sending ﬁve values of blood oxygen saturation every

AnApproachTowardsInformationQualityManagementofElectronic

HealthRecords

271

one minute and a listener expecting to receive it. Us-

ing the completeness quality dimension, we can de-

termine if the entire information package is received

by the listener or not.

As our system knows about the number of data

items in a package, it can evaluate the completeness of

the received information and assign a quality dimen-

sion value to it. Therefore, the completeness of an in-

formation package can be calculated utilizing Eq. (2)

in which N

received

shows the number of received data

items by the system and N

sent

represents the number

of sent data items from the device.

Completeness =

received

sent

∗ 100 (2)

2.1.3 Timeliness

The timeliness is the third proposed quality dimension

when generating EHRs. Timeliness refers to the fact

that the quality of an EHR can be sometimes strictly

associated to the time it has been generated. Some

EHRs must be committed to the system in a proper

time to be useful. For instance, you can think of a

notiﬁcation system in which EHRs are used to detect

the emergency condition of the subject.

In our systems, there is a time stamp for genera-

tion time and another one for the commitment time

of each EHR. Technically, if the difference between

these two times is more than the threshold, the EHR

is not meaningful. The threshold, cab be deﬁned ac-

cording to the level of importance of the time in our

system. For example, it is completely acceptable (but

not desirable) for an SMS to be delivered twenty four

hours after being sent but what about the report of a

heart attack? Typically in mobile networks, frequent

connections and disconnections are inevitable proper-

ties of the environment (Huang and Garcia-Molina,

2001). Therefore, the timeliness is very important

when determining the quality of an EHR.

The timeliness quality dimension can have vari-

ous deﬁnitions in different contexts. However, what

we generally need in this paper is a number (prefer-

ably between 0 and 100) which shows the quality di-

mension value. All we need to compute the timeliness

value, is to subtract the generation time (T generate)

from the commitment time T

commit

as it is shown in

Eq. (3).

Timeliness = T

commit

− T

generate

(3)

Table 1, explains how to convert the timeliness

value to the quality dimension values between 0 and

100. With the aid of this mapping table, the timeliness

becomes uniﬁed with other quality dimensions so that

we can calculate the ﬁnal quality metric.

Table 1: An example of timeliness mapping table.

T1 T2 T2 - T1 Timeliness

0 5 5 95%

0 10 10 90%

0 15 15 85%

Note that the timeliness dimension can be deﬁned

using other equations and mapping tables in respect

to the system environment. To design a mapping ta-

ble, the maximum value of the acceptable delay must

be deﬁned. This value is calculated considering the

importance level of the expected data, required time

for taking appropriate action, the type of network in-

frastructure (e.g. mobile, ﬁxed, etc.), and/or any other

inﬂuencing parameters.

2.1.4 Connectivity

Although connectivity is not among the popular qual-

ity dimensions, we believe that it can be useful in our

architecture. As mentioned earlier, frequent connec-

tions and disconnections of components are inalien-

able properties of mobile environments. Additionally,

mobile networks are very good environment for de-

ployment of our architecture. Therefore, connectivity

is very important and somehow it should be measured

and stored as metadata of EHR.

To make it more clear, assume that we have some

sensors measuring the heart pulse rate of a subject and

a system monitoring the values they generate (i.e. the

ECG curves). After a while, our monitoring system

receives no signal from the sensor. This incident, can

be interpreted at least in two ways: the loss of signal

or the subject’s heart attack. The loss of signal can be

caused by sensor failure, network issues, etc. but the

subject’s heart attack must be confronted differently.

The connectivity quality dimension demonstrates

the connection status of the network when an EHR is

being generated and committed. We assign 0 to the

connectivity dimension value when the sensor and/or

data gateway are disconnected from quality manager

and 100 when the connection is perfectly ﬁne.

2.1.5 Consistency

The last quality dimension we use is the consistency.

Consistency has a wide range of deﬁnitions in differ-

ent environments, but we use it as a quality dimension

which shows that whether EHR values are produced

in proper range of validity. In many health records,

especially those which are associated with measure-

ments, there is a range of validity. For example, in

EHR management systems there is a validity range

for heart pulse rate (e.g. 0-230).

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

272

Although the probability of experiencing a heart

rate out of this range is very low, it is possible to have

an anomalous value due to sensor failure. In such a

case, consistency is a very good dimension to deter-

mine the quality of the EHR. According to Eq. (4),

the consistency is calculated by the proportion of the

number of in-range data (N

in−range

) and the total num-

ber of data (N

total

Consistency =

in−range

total

∗ 100 (4)

2.1.6 Quality Metric

Now that we have all required quality dimension val-

ues, we can calculate the quality metric. Quality met-

rics are shown as numbers between 0 and 100 so that

bigger values show higher quality of the information.

In fact, the quality is an outcome of different qual-

ity metric. As mentioned, several different quality di-

mensions can be considered. The ﬁnal quality metric

is computed using Eq. (5) in which N is the number

of quality dimensions, (Q

) is the quality dimension

value and the (W

) is the weight of each quality di-

mension.

QualityMetric =

∑

i=1

∗Wi (5)

As you can see in Eq. (5), we deﬁne a weight

value (W

) for each quality dimension. These values

are used to control the effect of each quality dimen-

sion on the ﬁnal quality metric. For instance, in some

systems accuracy is more important than the timeli-

ness. In such a system we assign lesser weight to

the timeliness so that timeliness value can less affect

the quality metric in comparison with accuracy value.

This feature provides our system with more ﬂexibil-

ity and causes more compatibility to diverse environ-

ments.

3 QUALITY-ENHANCED EHR

APPROACH 1

Once we have calculated our quality metrics, we can

concentrate on our architecture. As shown in Figure

2, we propose an architecture in which a quality man-

ager is responsible for computing and assigning the

quality metric to the health record. Typically, a stan-

dard EHR consists some information about the pa-

tient, measured values, date and time, but we need

an EHR containing quality data to calculate the qual-

ity dimensions and quality metric. If a new system is

going to be developed, then these quality data can be

Figure 2: Quality-enhanced EHR management system.

easily considered in the system. But if we want to add

quality to an existing system, then extra information

must be added to the original EHR and it means that

some modiﬁcation in the data gateway is necessary.

Additionally, we need to modify the existing storage

space of EHRs.

As mentioned earlier, the majority of EHRM sys-

tems, use relational databases. Therefore, when

adding the quality metadata to the original EHR, mod-

iﬁcation of involved tables, views, stored procedures

and functions and other database objects must be per-

formed. Moreover, web/form interfaces which are

used for data entry should be updated as well.

When the data gateway sends quality-enhanced

data to the Q-EBS, the quality manager extracts the

quality data, calculates the quality dimensions and the

quality metric and stores them to the database. This

approach enables us to have the quality information

right beside the original EHR. Therefore, there is no

need to add any extra component to the current system

except the quality manager. The quality manager, can

be implemented as a single (while big) stored proce-

dure or function inside the database or an application

manipulating it.

4 QUALITY-ENHANCED EHR

APPROACH 2

As a major possible modiﬁcation of the ﬁrst approach,

we can design another independent component on the

sensor side which manages the generation of infor-

mation quality metadata. We name this component as

quality agent. The quality agent can accept various

roles in our architecture. For instance, it can just re-

trieve quality data from the data gateway and send it

to the quality manager in order to be processed. The

quality agent can also calculate the quality metric and

only send the results.

Figure 3 shows a possible scenario of using this

AnApproachTowardsInformationQualityManagementofElectronic

HealthRecords

273

component in our architecture. The principles of this

scenario are the same as what is shown in Figure

2, except that the quality agent component has been

added to it.

5 DISCUSSION

It is very important to decide on the kind of informa-

tion that should be sent to the quality manager since

this can affect the performance of the whole system.

Moreover, network trafﬁc, process distribution on the

smartphone and the EHRM server and the complex-

ity of implementation in term of updates propaga-

tion are directly associated with this decision. Table

2, demonstrates the effects of distribution strategies

on these aspects of the system, where trafﬁc means

the required bandwidth for communications between

quality agent and the quality manager, process means

the smartphone process load, and the number of up-

date refers to the number of necessary updates in case

of modiﬁcation of quality computation rules.

Table 2: The effects of sending quality data and quality met-

ric on various aspects of system performance.

Sent Item Trafﬁc Process No. of Updates

Q Data ↑ ↓ ↓

Q Metric ↓ ↑ ↑

There are some disadvantages with the quality

agent approach. However, it should be considered

as a possible option in mobile environments. Let

us brieﬂy describe some of its advantages and dis-

advantages. First of all, we must consider that hav-

ing a quality agent on sensor side (i.e. smartphone)

means that we have to completely split our architec-

ture into two sections: EHR management and infor-

mation quality management. It can result in a lot of

overheads in terms of process, communication, stor-

Figure 3: QEHRM system with quality agent.

age, security, etc. On the other hand this approach

enables us to manage the information quality in an in-

dependent way so that there will be no need to modify

the inner structure of existing systems. The only ma-

nipulation needed, is to store the metadata (i.e. qual-

ity dimensions and quality metric) in another database

with a link to corresponding EHR.

Secondly, computation of quality metric and then

sending it, may impose more process on the smart-

phone since the quality agent can be implemented

inside a smartphone. On the contrary, if the quality

agent calculates and sends the quality metric, then we

can decentralize the process and gain better network

efﬁciency since smaller data packets are transferred

through the network.

6 CONCLUSIONS

In this paper we proposed an architecture for assuring

the information quality of EHRs. Generally, our ar-

chitecture is based on ﬁve quality dimensions that can

be calculated using some metadata which are added

to the original EHR. We engaged four simple equa-

tions for calculation of EHR quality dimensions and

another one for computation of quality metric. Ad-

ditionally, we proposed two different implementation

alternatives for diverse environments. According the

section 5, the ﬁrst architecture is more suitable for

ﬁxed networks where the network bandwidth is not

our most important concern. Instead, we probably

prefer to distribute the process load of the centralized

server to the clients.

The second approach is more appropriate for mo-

bile networks where we want to use our valuable

bandwidth in the most efﬁcient way. Nowadays,

smartphones with high performance processors can

be found almost everywhere, so putting the burden of

some computation on smartphones seems trivial.

6.1 Future Work

As semantic web grows, more semantic-enhanced ap-

plications are developed and migrating from rela-

tional databases to ontologies becomes a must since

new customers have requirement that make conven-

tional approaches to face serious challenges (Berners-

Lee et al., 2001). Our proposed system uses some

statistical techniques to ensure the quality of informa-

tion. A magniﬁcent improvement of our system can

be the enhancement with semantic data and functions.

This can lead us to a system that can more precisely

detect the anomalous information and treat them in a

better way (Chandola et al., 2009).

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

274

Another component that can be used in our sys-

tem, is an intelligent EHR controller which is able to

improve itself to make better decision in case of fac-

ing abnormal records.

REFERENCES

Akyildiz, I. F., Su, W., Sankarasubramaniam, Y., and

Cayirci, E. (2002). A survey on sensor networks.

Communications magazine, IEEE, 40(8):102–114.

Baumgartner, N., Gottesheim, W., Mitsch, S., Retschitzeg-

ger, W., and Schwinger, W. (2010). Improving situ-

ation awareness in trafﬁc management. In Proc. Intl.

Conf. on Very Large Data Bases.

Berners-Lee, T., Hendler, J., Lassila, O., et al. (2001). The

semantic web. Scientiﬁc american, 284(5):28–37.

Chandola, V., Banerjee, A., and Kumar, V. (2009).

Anomaly detection: A survey. ACM Computing Sur-

veys (CSUR), 41(3):15.

DesRoches, C. M., Campbell, E. G., Rao, S. R., Donelan,

K., Ferris, T. G., Jha, A., Kaushal, R., Levy, D. E.,

Rosenbaum, S., Shields, A. E., et al. (2008). Elec-

tronic health records in ambulatory carea national sur-

vey of physicians. New England Journal of Medicine,

359(1):50–60.

Gunter, T. D. and Terry, N. P. (2005). The emergence of

national electronic health record architectures in the

united states and australia: models, costs, and ques-

tions. Journal of Medical Internet Research, 7(1).

Huang, Y. and Garcia-Molina, H. (2001). Publish/subscribe

in a mobile enviroment. In Proceedings of the 2nd

ACM international workshop on Data engineering for

wireless and mobile access, pages 27–34. ACM.

Kalra, D., Beale, T., and Heard, S. (2005). The openehr

foundation. Studies in health technology and infor-

matics, 115:153–173.

Seebregts, C. J., Mamlin, B. W., Biondich, P. G., Fraser,

H. S., Wolfe, B. A., Jazayeri, D., Allen, C., Miranda,

J., Baker, E., Musinguzi, N., et al. (2009). The open-

mrs implementers network. International journal of

medical informatics, 78(11):711–720.

Strong, D. M., Lee, Y. W., and Wang, R. Y. (1997). Data

quality in context. Communications of the ACM,

40(5):103–110.

Wand, Y. and Wang, R. Y. (1996). Anchoring data quality

dimensions in ontological foundations. Communica-

tions of the ACM, 39(11):86–95.

AnApproachTowardsInformationQualityManagementofElectronic

HealthRecords

275