Connections of Reduced Performance Health Data for Severe

Persistent Uncontrolled Allergic Asthma Treated by Omazulimab

Stefanos Matsopoulos and Valentina Plekhanova

Department of Computing, Enineering and Technology, University of Sunderland, Sunderland, U.K.

Keywords: Association Rules, Asthma, Data Mining, Spirometry Data.

Abstract: An application of association rules mining method for the discovery of associations in abnormal quantitative

health data for inadequately controlled severe allergic Asthma treated by Omazulimab is presented. To the

best of authors’ knowledge, no formal approaches have ever been used for extraction of association rules

among dysfunctional elements in Spirometry datasets. Initially is provided an explanation of the procedures

used for diagnosing inadequately controlled severe allergic Asthma. Following this, it is conducted critical

evaluation of well-known ‘association rule’ mining techniques, in order to identify the one with the best

utility for discovery of associations among abnormal elements of Spirometry datasets. Apriori Algorithm is

applied to real-life Spirometry datasets to illustrate the contribution of application of association rule mining

techniques. This revealed the existence of association rules among dysfunctional Spirometry elements for

this disease. Moreover it has been identified that this disease is provoked by association of Spirometry

elements that do not function properly as these are provided by Spirometer. This is translated in human

factors as a dysfunction of small and medium airways of patients’. Furthermore Spirometry element FEV1,

is not as valuable parameter as the European Medical Agency supports. Finally it has been observed that

Omazulimab treatment improves respiratory function and makes the connection among associated elements

weaker.

1 INTRODUCTION

Asthma is a disease that affects a large portion of

global population. Although asthma is separated into

different categories according to its severity, with

the most dangerous type being ‘poorly controlled

severe allergic Asthma’. This type of Asthma may

have mortal effects if not treated, or if patients do

not get the proper treatment in time, which occurs

only after hospitalization (Bousquet et al, 2007).

Immunoglobulin E (IgE) is a responsible factor

for severe allergic Asthma. During the last decade a

treatment has been discovered, called Omazulimab,

which is able to stabilize IgE. The appropriate

dosage is calculated by a practitioner and its use

leads to the reduction of exacerbations and

hospitalization (Nowak, 2006). Furthermore this

treatment can improve the patients’ Quality of Life

and reduce the risk of effects caused by Asthma

(Nowak, 2006).

The difference between inadequately controlled

severe allergic Asthma with other types of Asthma is

its severity. It has plenty of exacerbations and it is

not treated with the same medication as other types

of Asthma. The use of Inhaled Corticosteroids (ICS)

and Long-Acting-b2-Agonistics (LABAs) are not

effective on this type of Asthma. It has to be treated

with Omazulimab which has the ability to reduce

high number of free-IgE that patients suffer from

(Tzortzaki et al, 2012).

For an individual to be evaluated by a

practitioner as a patient who requires treatment with

Omalizumab, there are minimum medical

requirements that have to be met. These will be

discussed below. An important requirement that has

to be fulfilled is an abnormal FEV1 value (air

exhaled in the first second) (Tzortzaki et al, 2012).

Spirometry is a significant examination for the

measurement of lung function. It is used to measure

patients’ respiratory health and is helpful in

assessing conditions such as Chronic Obstructive

Pulmonary Disease and Asthma (Stout et al, 2012).

Spirometry provides an outcome which consists

of several numerical parameters. Each parameter

represents functionality of a different part of the

human respiratory. Also this examination is used to

276

Matsopoulos S. and Plekhanova V..

Connections of Reduced Performance Health Data for Severe Persistent Uncontrolled Allergic Asthma Treated by Omazulimab.

DOI: 10.5220/0004763802760286

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2014), pages 276-286

ISBN: 978-989-758-010-9

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

measure response to treatment of conditions which

Spirometry detects (Stout et al, 2012).

In general results above 80% predicted,

according to the ideal, are considered normal, while

results under 80% are considered abnormal and will

be discussed below.

Spirometry data are collected and used for the

discovery of associations of abnormal Spirometry

results/values. This paper is based on the analysis of

data through Data Mining Techniques. These

techniques require quantitative-numerical data for

concluding to a result. These are given by

Spirometry examinations, which are the only

examinations that provided quantitative data during

treatment. They are used for representation of

patients’ clinical image and after appropriate

analysis with ‘data mining’ techniques will lead to

associations of abnormal Spirometry parameters

(Nowak, 2006).

Except Spirometry data analysis, further analysis

of health examinations by practitioner has to be

taken into consideration when dealing with

inadequately controlled severe allergic Asthma. This

action leads to higher diagnostic accuracy. Forthwith

the problem will be approached and examined only

by a quantitative point of view and not overview of

all health examinations. The reason for approaching

this subject by a different point of view is because it

has been found that Omalizumab treatment is not

effective on all patients for unclear reasons (Slavin

et al, 2009). After appropriate management and

analysis of Spirometry data, this paper focuses on

‘hidden’ information which exists beneath

Spirometry examination results and cannot be

analysed in detail without an appropriate technique.

In order to understand the nature of the problem,

relevant information and techniques that are being

used in traditional diagnosis and treatment

procedures are required to be briefly introduced for

the disease.

The structure of this paper is as follows. In the

second section key aspects of severe allergic

Asthma, with poorly controlled symptoms are

analysed as are the methods used for diagnosing and

evaluating Spirometry results.

Moreover in the third section there is a critical

evaluation of data mining techniques. In the fourth

section characteristics of the sample and simulation

of the technique used are presented. Finally the last

section focuses on recommendations for appropriate

use of this technique.

The following research questions will be

addressed in this paper:

Research Question 1: Are there any ‘data

mining’ techniques that can find ‘association rules’

in Spirometry Data?

Research Question 2: Is there a ‘data mining’

technique that could be useful in the discovery of

associations among Spirometry parameters for

severe allergic Asthma inadequately controlled?

2 BACKGROUND

OF INFORMATION

MANAGEMENT IN ASTHMA

DISEASE: KEY ASPECTS

As defined by multiple authors Asthma is an

inflammatory disorder of the lung-function and its

airways. It involves multiple inflammatory cells and

the interaction of a big amount of different

mediators of the lungs (Holgate et al, 2009).

Asthma is divided into mild and moderate, that

will not be further analysed in this paper and severe

persistent allergic Asthma, with inadequately

controlled symptoms. (Korn et al, 2009).

2.1 Severe Allergic Asthma

Inadequately Controlled: Key

Aspects

Severe allergic Asthma with inadequately controlled

symptoms can be characterized by the presence of

IgE antibodies against allergens that are common

such as animal dander or house dust (Slavin et al,

2009).

This can lead to lack of oxygen and incorrect

lung function leading to hospitalisation or increased

risk of death. (Nowak, 2006). Furthermore there are

multiple exacerbations during a small period of time

that can decrease the patients’ Quality of Life.

The use of Omazulimab treatment, which

contains Anti-Immunoglobulin E (anti-IgE), is

useful for patients’ who suffer from this type of

Asthma (Tzortzaki et al, 2012).

Though it is unclear why the use of Omalizumab

is not effective on all patients (Slavin et al, 2009).

The relationship between clinical symptoms and free

IgE has not been studied adequately. Consequently it

is difficult to predict the characteristics of the

patients that will be positively affected by the

treatment (Holgate et al, 2009).

2.2 Minimum Medical Requirements to

Initiate Omazulimab Treatment

For an individual to be taken under consideration for

ConnectionsofReducedPerformanceHealthDataforSeverePersistentUncontrolledAllergicAsthmaTreatedby

Omazulimab

277

Omazulimab treatment some minimum medical

requirements have to be met. Patients have to be

older than 12 years old and have a positive skin

prick test or have reactivity to a perennial

aeroallergen (Korn et al, 2009). Moreover lung

function and especially FEV1 has to be lower than

80% as supported by the European Medical Agency

(Korn et al, 2009). Additionally the number of

severe asthma exacerbations that individuals may

suffer from, no matter what amounts of ICS and

LABAs is being used, has to be observed. Finally

patients must have a treatable IgE over 30 IU/ml,

which is measured before treatment starts.

Nevertheless the most important factor is the

practitioners’ evaluation on the patients’ overall

health and their need for treatment.

2.3 Utility of Spirometry in Severe

Allergic Asthma Inadequately

Controlled

As discussed above, one of the minimum

requirements for receiving Omazulimab treatment is

maintaining a FEV1 percentage less than 80% of the

‘normal’ measurement. ‘Normal’ is calculated by a

Spirometer based on age, gender, height and weight

(Stout et al, 2012). Every 2 to 4 weeks, depending

on patient’s dosage a Spirometry examination is

performed (Holgate et al, 2009). With this

examination a practitioner will be able to evaluate

the patient’s improvement. Furthermore Spirometry

is a ‘mirror-indicator’ of a patients’ respiratory

function and cannot be examined by any other

quantitative test (Stout et al, 2012).

2.4 Traditional Procedures on

Diagnosis of Improvement during

Treatment

Traditionally practitioners do not have a specific

technique which can be used for accurate

identification and evaluation of patients’ health.

Instead there is a procedure that has to be followed.

After evaluation and comparison of different

examinations a decision is made based only on the

practitioners’ opinion. There is no relevant technique

for an in-depth analysis of the examination results.

Forthwith, studies show that after a number of

exacerbations (Nowak, 2006) Spirometry

examination is the most important factor for the

evaluation of respiratory performance.

Due to the fact that IgE is examined only before

treatment starts, Spirometry is the only examination

that can reveal the patients’ lung function and

overall respiratory health (Bousquet et al, 2007).

Also based on previous research, (Tzortzaki et al,

2012) the only factors that are important and are

taken into consideration are Forced Vital Capacity

(FVC), Forced Expiratory Volume 1 (FEV1) and

Peak Expiratory Flow (PEF). However not all

authors take into account PEF measurements, for the

evaluation of patients’ respiratory function

(Tzortzaki et al, 2012). Although the research of this

paper will not only focus to only to these three

elements. Contrariwise in this paper PEF factor and

all factors that are provided by a Spirometry

examination will be considered.

A diagnosis is made by a practitioner and

improvement of respiratory performance is

evaluated. This occurs after a procedure that entails

the evaluation of a patients’ health, the number of

exacerbation and the percentage of FEV1 and FVC

according to the ‘normal’ measurement which is

provided by a spirometer.

Previous research has shown (Korn et al, 2009)

that secure and accurate measurement of results can

be assessed after at least 16 weeks of treatment. On

the other hand only 12 weeks of treatment can create

a secure overview of treatments’ outcomes (Slavin et

al, 2009).

2.5 The Proposed Scenario for

Examination of Spirometry

Elements

In this paper we consider the following scenario for

examination of Spirometry elements: each patient

has a number for Spirometry examinations taken on

different dates. Each Spirometry examination

consists of elements which measure parts of the

respiratory system. Elements such as FEV1

represent the percentage of function of different part

of human respiratory, according to the ‘ideal’

function that each individual should have. The

‘ideal’ is created based on each individual personal

characteristics, which will be analysed below. All

abnormal (parameters) elements, which will be

analysed below, will be added to the proposed

technique. After simulation of the technique to the

Spirometry data a combination of abnormal

elements (parameters) will be revealed.

This procedure approaches each patient

separately, as a different database. Each patient

provides an association of dysfunctional elements.

Following a statistical analysis is taking place which

leads to associations of abnormal Spirometry

elements that are met to most of patients. The

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

278

following issues could be addressed: patients could

have different number of days for examinations;

different Spirometry elements but some could be the

same as in the previous examination days.

3 ASSOCIATION RULES

AND SPIROMETRY DATA

Quantitative data that are relevant to measurement of

Spirometry elements, are used for a consideration of

examination results. They are divided into before

and during a treatment period. The results are then

further analysed in order to make the final judgment

on the patients’ health improvement. This work

examines Spirometry data to find if there is an

association among Spirometry abnormal parameters.

Identified associations could be used for better

diagnosis and/or Asthma prediction.

This section presents well known techniques of

‘data mining’ which are used in analysis of

associations from the perspective of application to

Spirometry data. ‘Association analysis’ is used for

discovering interesting or uncovered relationships

hidden in data sets. ‘Association rules’ are used for

data of multiple nature/types and each one of them

has different criteria for significance of outcomes

(Lee et al, 2005), (Guang-Yuana et al, 2011).

Let us define the following key aspects: let I =

, i

,…, i

} be the set of all Spirometry elements,

e.g. Forced Expiratory Flow when 50% and 25% of

total air are blown (MEF 50, MEF 25 respectively),

Forced Expiratory Flow between 25 percent and 75

percent of the Vital Capacity (MMEF 75/25), FEV1

(i.e. items) in a Spirometry dataset for one patient

during examination period of time and T = { t

, t

,…, t

} be the set of examination results relevant

to Spirometry data elements for one patient (i.e.

“transactions”). Each Spirometry examination ti

contains a subset of k-items (i.e. a k- itemset) chosen

from I.

We are interested in associations of abnormal

Spirometry elements and we seek to find association

rules that define an implication A → B, where A is

MEF 50 and B is MEF 25. A and B are disjoint

itemsets, i.e., A∩B=Ø.

The strength of an ‘association rule’ can be

measured in terms of its ‘support’ and ‘confidence’.

‘Support’ determines how often a rule is applicable

to a given i.e., patient Spirometry dataset, while

‘confidence’ determines how frequently items in B

appear in Spirometry dates that contain A. A

‘support’ measure in Spirometry data identifies rules

that have very low occurrence on examination

results. For this reason, ‘support’ can be used to

eliminate uninteresting rules in Spirometry data. On

the other hand ‘confidence’ can be defined as the

probability of element/item A to be found at the

same time in combination with element/item B,

where A and B are Spirometry elements such as

MEF 50 and MEF 25 respectively. Both ‘support’

and ‘confidence’ measures have to be equal or

bigger than the one that ‘user’ has specified as

minimum. Otherwise items and combinations of

them are not taken into account (Lee et al, 2005).

Furthermore ‘confidence’ measures the reliability of

the inference/implication made by a rule. For a given

‘association rule’, the higher the ‘confidence’, the

more likely it is for an elements/items such as

Spirometry elements to be present in a patient

Spirometry dataset, that contains elements such as

MEF 50 and MEF 25. That is how, ‘confidence’

determines how frequently items in

transactions/Spirometry dates, appear in a patient

Spirometry dataset that contains dysfunctional

elements.

Subsequently, three ‘association rules’

techniques are further analysed. The reason for this

selection is because they are the most well-known

and used ‘association rules’ techniques (Cokpinar

and Gundem, 2012). These are: Apriori Algorithm,

Frequent Pattern (FP) Growth and Sampling

algorithm. Each one has advantages and

disadvantages, which are further discussed, although

it is highly important to adopt the nature/type of

Spirometry data and the nature of the problem for

the production of a significant outcome.

3.1 Apriori Algorithm

Apriori Algorithm is a procedure that scans the

frequent Itemset of Spirometry data to reveal

associations that satisfy minimum ‘support’.

Consequently this algorithm generates new

Candidates and re-scans the initial database to find

more associations among elements/items at one at a

time/scan (Yu et al, 2010). Finally it terminates

when no more associations/combinations can be

found among elements. For example it checks only

for element A if it satisfies minimum ‘support’ then

for B, C and so on. In its second scan it examines for

elements that have passed minimum ‘support’ in the

first step only. It generate combinations of A and B,

A and C and it continues for all elements of the

sample, where A, B and C are elements such as

MEF 50, MEF 25 and MMEF 75/25 respectively. In

a new candidate combinations of two elements that

ConnectionsofReducedPerformanceHealthDataforSeverePersistentUncontrolledAllergicAsthmaTreatedby

Omazulimab

279

satisfy minimum ‘support’ are stored. In its next

scan it searches for combinations of 3 elements and

stores them again. This procedure continues until no

more combinations of abnormal elements that satisfy

minimum ‘support’ can be found. Consequently all

Itemsets that do not satisfy minimum ‘confidence’

are abandoned. Its final outcome is a set of

combined Items that fulfil minimum requirements.

One of its great advantages is that it has great

attention to detail and increases accuracy

accordingly to sample size. Improved accuracy can

be achieved to bigger samples. (Ykhlef, 2011).

One of the drawbacks of the Apriori Algorithm

that has been observed by researchers is (Umarani

and Punithavalli, 2011) the need of an entire Dataset

scan for each item added on the Itemset. This results

in slow implementation and high execution time.

Moreover because of its delay more electricity is

needed and because of technical equipment such as

Computers and storage space there is an increase of

cost (Liu et al, 2012). Another problem that is

encountered is the need for additional storage space

and this is due to the creation of new Itemsets.

Storing is significantly important for future analysis

and evaluation through Apriori. As a result,

acquisition of technical equipment (Computer,

Storage space) will lead to an additional cost and

may be unaffordable. A basic influential factor is the

size of the Itemset that is used for analysis (Nahar et

al, 2012). This problem is also encountered if there

is a small ‘support’, set by the ‘user’, which will

lead to the creation of multiple new Itemsets

(Umarani and Punithavalli, 2011). Consequently, the

smaller the sample the faster and cheaper the

implementation will be.

3.2 Frequent Pattern Growth

Algorithm

In Frequent Pattern Growth Algorithm the procedure

to solve the problem is divided into two steps

(Umarani and Punithavalli, 2011). In the first step

the database is scanned only once and an FP-tree is

created. In each branch of FP-tree Items

/Spirometry elements from dataset/Spirometry

examination results from the dataset are added and

stored by their names. In every new element a new

sub-branch is created and stored with a Transaction

ID (TiD). A unique ‘path’ of each transaction is also

created. The size of FP-tree is at least as a big as the

original Database.

In the second step it recursively mines all the

patterns from the FP-tree that was built in the first

step and concludes on the result. It follows the

prefixed path based on a specific search e.g. path

that contains a specific element as FEV1 does. This

leads to results that satisfy ‘support’ by exclusion of

TiD that do not satisfy ‘support’ and ‘confidence’.

The creation of smaller structures (FP-tree), if

not the same size as the entire database, is analysed

(Chen et al, 2011) is one of the advantages of FP-

Growth Algorithm. Also it creates a methodology

that does not demand the creation of multiple

Candidate Itemsets. Finally it reduces the search

space that is needed because of its methodology.

Therefore only two patterns of development are

needed, with smaller execution time (Lin et al,

2011). This results in faster analysis of Spirometry

examination of patients who have large amount of

examinations.

One of its drawbacks is that it needs the same

amount of storage and memory capacity as Apriori

does (Liu et al, 2012). It also has high cost because

of high demand of technical equipment (Computer,

Storage space, Electricity). Moreover there are

crossing authors opinions about Aprioris’

implementation speed (Ke et al, 2013), (Umarani

and Punithavalli, 2011). Meanwhile it has high

efficiency; its complexity in each step (Duneja and

Sachan, 2012) results in the presence of a great

barrier for individuals/practitioners who lack the

experience in modelling aspects; to calculate

findings and execute proper analysis of data.

Finally every new Spirometry dataset of each

patient is added to the previous database. Some

patients may have thousands of Spirometry

examinations which could lead to a massive

Database. As it has been discussed, FP-growth is

based on the development of a FP-tree. This leads to

bigger samples which have reduced mining

performance (Lin et al, 2011). Thereafter there are

some patients with a lot of Spirometry datasets.

Lower mining performance which can lead to lower

significance of results is not desirable for nature of

Spirometry Data and Health research that needs high

accuracy.

3.3 Sampling Algorithm

Sampling algorithm does not have a specific

structure as Apriori and FP-growth algorithms do.

Instead it is based on a different algorithm, it is

based on the nature of Data. By appropriate

algorithm it generates an accurate sample size for

implementation initially. It provides an outcome

significance and accuracy of sample size. If it is not

appropriate it is modified and repeats this step. After

identification of appropriate sample size, the final

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

280

outcome is provided. Analysis of the sample takes

place by other Association Rules techniques such as

Apriori or FP Growth Algorithm (Chen et al, 2011).

As is revealed by its name it uses a sample of the

entire database and analyses it (Umarani and

Punithavalli, 2010).

Its methodology is divided into a number of steps

that are depended on the algorithms that will be used

for implementation.

Initially the selection of the sample takes place

during step one (Umarani and Punithavalli, 2010).

The selection of the sample that leads to accurate

results differs in each type of data. The most

efficient technique can be revealed only after its

implementation. A random selection from the

database can be found if it satisfies ‘support’ by

implementing it by giving a slightly lower minimum

‘support’ than the one that the ‘user’ specified (Chen

et al, 2011). If it is not satisfied then more Data are

being added to the sample and the sampling

technique is re-implemented. The procedure

terminates when the sample satisfies minimum

‘support’ of the ‘user’.

Consequently the sample is analysed by an

‘association rule’, like Apriori or FP-growth and

results are provided after implementation (Ackan et

al, 2008).

Advantages that have been identified are the

decreased use of storage space, the low cost because

of the lower demand of technical equipment

(computer, storage space). Moreover it is

implemented faster than the total analysis of the

database because less Data have to be analysed

(Umarani and Punithavalli, 2010).

Each algorithm has particular strengths and

weaknesses, relevant to the nature/type of data that

are being used. The same rule applies for the

Sampling Algorithm on Spirometry data. Although

the Sampling algorithm has disadvantages on its

own thus the possibility of data that the sample is

consisted to be similar. This could result in faulty

outcomes of the analysis. Moreover quality and

accuracy of outcome depends on the database size.

Databases of bigger size have higher accuracy and

better quality than smaller ones. This is not

appropriate and effective on nature/type of

Spirometry data because some patients have plenty

of examinations and others have few.

3.4 Critical Comparison of Techniques

in Relation to Data Needs

Apriori and FP-Growth have high efficiency, quality

of outcome and pay attention to detail. These

advantages do not apply to the Sampling algorithm.

On the other hand they have high cost and long

implementation time, which can lead to delayed

results.

For an outcome of high significance medical data

need to have accuracy of results. As a result, the

most important factor for the selection of the most

appropriate algorithm is based on its accuracy,

quality and attention to detail of the data. Each

sample method needs to be implemented in order to

find the most suitable for Spirometry type of data.

Based on these requirements, Sampling Algorithm is

excluded because it lacks of these requirements.

Moreover, the need for mathematical and data

analytical knowledge for implementing the sample

algorithm process, are additional reasons for not

selecting this algorithm for the analysis of this type

of data.

Likewise FP-growth will not be selected for

analysis of Spirometry data because of its high

complexity during implementation. Furthermore

exceptional mathematical and ‘data mining’

knowledge is required by the ‘user’. Also it has

similar cost as Apriori.

Concluding based on the above comparisons, the

best suited algorithm that will be used for analysis of

our Data is the Apriori algorithm. The reason for this

decision is that it fulfils limitations such as accuracy

and quality of outcome in better significance than

the FP Growth and Sampling algorithm. Moreover

the ability to analyse big Datasets, based on size, it

affects positively the outcome (Amato et al, 2011).

The larger the sample the more accurate the

outcome. Also it is important to the ‘user’ because

multiple stored data (Transactions) from previous

years are needed for analysis. So as time passes, the

sample becomes bigger. Furthermore it can be

adopted by Spirometry Data with no limitations.

The Apriori algorithm is used as a guide for

finding associations among elements of Spirometry

data that are dysfunctional/abnormal (<80%) such as

MEF 50 and MEF 25. Additionally ‘support’

measure helps to prune candidate Itemsets

discovered during frequent Itemset generation on

Spirometry dates/transactions.

ConnectionsofReducedPerformanceHealthDataforSeverePersistentUncontrolledAllergicAsthmaTreatedby

Omazulimab

281

4 APPLICATION OF APRIORI

ALGORITHM TO REAL-LIFE

SPIROMETRY DATA SAMPLE

4.1 Sample Characteristics of Patients

All Data were collected from the General Hospital

of Ioannina ‘Hatzikosta’ in Greece. The sample that

was chosen for analysis consists of 20 patients; 13 of

which are women and 7 are men. All patients’ are

residents in the region of Epirus and are under

treatment with Omazulimab in the General Hospital

‘Hatzikosta’. This sample of 20 patients is consisted

by 559 Spirometry examinations that were taken for

analysis; 12 of which (2, 14% of total sample) have

not been taken into consideration for analysis due to

the lack of total Spirometry elements indications.

This was caused by mechanic failure of the

Spirometer. Moreover examinations before the 12

week of treatment are not being taken into

consideration as explained below (10, 7% of total

sample). Furthermore each patient had a different

number of examinations. The average number of

examinations per patients is almost 28 (27, 9 is the

exact number) with the minimum number of

examinations being 9 per patient and maximum

being 60 per patient.

Moreover the average IgE level of our sample is

535, 65 with lowest being 140 and the highest 1476.

Also the average age of our patients was 64 (63, 8)

with youngest being 48 years of age and oldest being

78. Finally the average treatment period of sample

was almost 21 months (21, 15 is the exact number)

with shortest being 7 and longest period being 40

months per patient. Although averages have been

computed for the total sample, a portion of our

patients (10 out of 20) are still under treatment and

others have completed the treatment. Finally the

examinations that have been collected are from July

of 2001 until July of 2012. Finally 10 out of 20

patients have completed the Omazulimab treatment.

So ‘after treatment’ period examinations were not

further analysed for higher significance of results.

4.2 Process Followed for Analysis of

Spirometry Data based on Study

and Disease’s Requirements

Data were collected from 20 patients in three

periods: Spirometry examination before

Omazulimab’s treatment, after 12 weeks of

treatment and after treatments’ completion. The first

period consists of Spirometry examinations that

have been taken before Omazulimab’s treatment, 31,

3% of the sample (175 out of 559) as shown in Table

1. The percentage of each element, on each patient

who has not proper function is presented.

The second period is composed by Spirometry

examinations taken after 12 weeks of treatment until

the completion of the treatment and represents 51,

8% of total sample. The effectiveness of

Omazulimab is visible and proves the previous

researches (Slavin et al, 2009). In the second table it

is presented ‘during treatment’ period and shows the

percentage of examinations that each Spirometry

element is abnormal (>80%).

The last period consists of Spirometry

examinations taken after the treatments’ completion.

Although this period is not taken into consideration

for analysis due to lack of examinations, as only

50% of the sample (10 out of 20 patients) has

completed the treatment. Something that is

translated to 6, 08 % of total sample examinations

(34 out of 559).

Each element given by Spirometry as healthy

and functional has to have at least 80% function

performance for the ‘ideal’ measurement (National

Heart Lung, and Blood Institute, 2007). This is given

for each element separately by the Spirometer. Only

elements that fulfilled this requirement were taken

into consideration. The identification of associations

among elements that are not fully functional and are

responsible for this disease is being researched.

Information about each patient is presented by a

dataset, from where associations of elements

occurred after analysis of Spirometry examinations.

Each date of a Spirometry examination was set as a

Transaction and elements in it were set as Items.

Below 80% were taken into consideration (set as 1)

and above 80% were not (set as 0). Following this

allocation each patients’ Spirometry examinations

(dataset) have been imported and analyzed with

Apriori. The process that has been followed is

briefly described in section 3.1. After this analysis

combination of abnormal Spirometry parameters for

each patient has been discovered. Each patient has

been analysed twice as different Datasets for before

and during treatment period were used.

The provided technique has only been used as an

additional technique for practitioner to make more

accurate diagnose and evaluation patients’ health on

top of other Medical findings.

‘Support’ measurement has been set to 66, 6%

for having significance of result, as abnormal

elements have to be more of a half times in

examinations sample. Additionally ‘confidence’

measurement has been set to 100% because the

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

282

Table 1: ‘Before’ treatment period.

Patients

Number

FEV

FVC

FEV 1 %

VC MAX

MEF

PEF

MMEF

75/25

FVC

Out

1 22.72 22.72 63.63 9.09 22.72 14 100 100 22.72 100 22.72 22

2 0 0 0 0 0 0 100 100 0 100 0 2

3 4.76 38.09 28.57 28.57 4.76 66.66 100 85.71 57.14 100 0 21

4 37.5 37.5 37.5 37.5 12.5 50 75 100 25 75 37.5 8

5 50 50 50 50 25 50 75 100 50 75 50 4

6 78.94 94.73 100 94.73 31.57 100 100 100 100 100 78.94 19

7 12.5 0 0 0 0 50 100 100 25 100 12.5 8

8 50 0 100 0 0 50 100 100 50 100 50 2

9 0 0 0 0 0 0 100 100 0 100 0 1

10 0 0 33.33 0 0 33.33 100 100 0 100 0 3

11 38.46 30.76 38.46 26.92 19.23 65.38 61.53 80.76 42.3 69.23 38.46 26

12 0 0 0 0 0 0 100 100 0 100 0 1

13 0 0 0 0 0 77.77 66.66 55.55 66.66 66.66 11.11 9

14 50 50 50 50 0 100 100 100 100 100 25 4

15 50 50 50 25 0 25 50 50 0 75 50 4

16 5 5 5 0 0 30 80 70 15 85 5 20

17 100 100 100 100 100 0 100 100 100 100 100 3

18 66.66 33.33 33.33 33.33 0 33.33 66.66 100 33.33 66.66 66.66 3

19 100 100 100 100 40 100 100 100 100 100 100 5

20 10 20 10 20 0 80 100 100 70 100 10 10

Table 2: ‘During’ treatment period.

Patient

Number

EX FEV1 FVC

FEV1 %

VC MAX

MEF

25 PEF

MMEF

75/25

FVC

Out

1 10 20 30 0 0 50 100 100 30 100 20 10

2 0 0 0 0 5 15 70 100 0 90 5 20

3 0 20 20 20 0 40 100 100 60 100 0 5

4 100 100 100 100 9.09 100 100 100 27.27 100 100 11

5 14.29 14.29 14.29 14.29 0 14.29 100 100 14.29 100 14.29 7

6 100 100 100 100 0 100 100 100 100 100 100 6

7 0 0 0 0 0 0 100 100 0 100 0 1

8 54.17 54.17 87.50 45.83 0 66.67 100 100 8.33 100 54.17 24

9 23.53 0 0 0 0 0 23.53 88.24 0 58.82 23.53 17

10 6.25 3.1 9.38 3.13 0 21.88 46.88 96.88 18.75 71.88 12.5 32

11 0 0 5.56 0 0 38.89 77.78 88.89 0 88.89 0 18

12 0 0 0 0 0 100 100 100 0 100 0 5

13 0 0 0 0 0 100 100 100 83.33 100 0 6

14 0 31.,03 27.59 13.79 0 100 100 100 93.10 100 10.34 29

15 0 0 0 0 0 0 0 100 0 0 0 4

16 0 2.78 0 0 0 44.44 94.44 100 33.33 97.22 0 36

17 100 100 100 100 0 100 100 100 100 100 81.82 11

18 37.5 31.25 50 25.00 0 93.75 100 100 0 100 43.75 16

19 86.67 100 100 86.67 86.67 100 100 100 100 100 86.67 15

20 5.88 0 5.88 0 0 70.59 100 100 29.41 100 5.88 17

ConnectionsofReducedPerformanceHealthDataforSeverePersistentUncontrolledAllergicAsthmaTreatedby

Omazulimab

283

nature of data demands high accuracy.

Although 20 associations occurred (one of each

patient/Dataset) only patients that met minimum

‘support’ and ‘confidence’ measurement criteria

were statistically evaluated. And separately for each

period which led to associations of Spirometry

elements that are mostly met among patients.

Patients that failed to meet the required minimum

criteria were not considered in the process of

statistical analysis for achieving improved accuracy

and significance of final results.

Finally a comparison of associations/

combinations, before and during treatment periods

has been made. This led to an outcome which

revealed the amount of associations of ‘before

treatment’ period that have been healed by the use of

Omazulimab in ‘during treatment’ period.

4.3 Discussion of Results from Patients’

Spirometry Data Analysis

It has been found that in ‘before’ treatment period

there were three patients who although satisfied the

‘support’ indicator through the full process of

analysis, the association of Spirometry

parameters/elements did not fulfil ‘confidence’

indicator. This resulted in the exclusion from the

second step of analysis which is the Statistical

analysis of the sample. Also one patient had only

one element that fulfilled ‘support’ indicator. This

concluded to an end of his/her analysis because there

were no associations to be found. Additionally the

patient has also been excluded from second step of

analysis.

An association of elements that has been found

to 50% (8 out of 16) of sample that satisfied

‘support’ and ‘confidence’ with no other elements

on it, is consisted by the elements MEF 50, MEF 25

and MMEF 75/25. From which 75% (6 out of 8) had

100% Support. Also this combination has been

found in associations with other elements to 93, 75%

(15 out of 16) of sample. Also it has satisfied

Support and Confidence limitations.

In ‘during’ treatment period two patients had

only one element that fulfilled ‘support’ indicator.

This concluded to end of analysis because there

were no associations to be examined. Also they were

excluded from second step of Statistical analysis.

The same association/combination of elements

has been found in ‘before’ and in ‘during’ treatment

period. It is consisted by MEF 50, MEF 25 and

MMEF 75/25 has been found to 44, 45% (8 out of

18) of sample. It also satisfies ‘support’ and

‘confidence’ limitation. Although it has been found

at the same height as ‘before’ treatment period 37,

5% (3 out of 8) of patients are different. Furthermore

it has been found that associations that have 100%

‘support’ have been reduced to 62, 5% (5 out of 8).

Furthermore this combination is met in associations

with other elements to 94, 44% (17 out of 18) of

sample that satisfied ‘support’ and ‘confidence’.

Another significant finding that has been

revealed from this work is the importance of FEV1

for diagnosis of this disease. It has been revealed

that although FEV1 is an element that is a

prerequisite medical requirement for beginning of

treatment, as also a significant measurement for

evaluation of it, it is not as dysfunctional as it had

been initially supported. The reason of this outcome

is because it has not fulfilled ‘support’ indicator to

most of patients neither ‘before’ nor ‘during’

treatment period. Although it has been met to 25%

(4 out of 16) of associations on ‘before’ treatment

period and in 22, 22% (4 out of 18) to ‘during’

treatment period, it has not significant appearance. It

was expected to be found to all datasets and to be

highly significant as it is one of the minimum

Medical requirements to begin Omazulimab

treatment according to European Medical Agency

(Tzortzaki et al, 2012)

Furthermore 100% ‘confidence’ had been

demanded because of Data nature. It has been

indicated by some factors that there is a possibility

for some Spirometry elements/parameters with

lower than 100% ‘confidence’ to fulfil needs of

analysis. It has been found to some elements and

there is high possibility to indicate individuals that

suffer from severe allergic Asthma inadequately

controlled and it is not reducing significance of

finding.

4.4 Critical Evaluation of the

Outcomes

After implementation of Apriori Algorithm and

analysis of sample, we identified an association of

elements. It is the same in ‘before’ and ‘during’

treatment period. The association that is mostly on

both periods is consisted by MEF 50, MEF 25 and

MMEF 75/25 elements. Importance of results is

being provided in more detail in section 3, where

‘Support’ and ‘Confidence’ measures are discussed

in the context of their application to Spirometry data.

The meaning of this finding in health value is that

when a patient has low flow speed of air moving out

of his lungs at the time of 50% of FVC is blown it

has also low flow of air blown when 25% of FVC is

blown and his/her small and medium airways are not

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

284

as functional as they had to be based on the ‘ideal’.

Patients who fulfil other medical requirements

are indicated by this association of elements as

patients who require treatment with Omazulimab.

The reason is that they suffer or will suffer in future

by severe allergic Asthma inadequately controlled.

It has been found the same association of

elements to both periods and revealed that treatment

has a positive effect on this association of elements.

Level of dependence among these elements became

weaker. Elements are not that close correlated after

12 weeks of treatment or more. Furthermore it has

been proved that treatment has positive reaction

except medical view point also by Spirometry Data

point of view.

Encounter to less than 25% of sample, on both

periods, of FEV1 has reduced its utility as a

minimum medical requirement for start of

Omazulimab treatment.

Furthermore associations of elements that

provoke severe allergic Asthma inadequately

controlled are MEF50, MEF25 and MMEF75/25.

This association/combination of Spirometry

elements it has been found as a ‘stand-alone’

combination. Although it has been met in

associations with other elements in both ‘before’ and

‘during’ treatment periods.

In addition it has been revealed that Omazulimab

treatment has positive reaction upon this association

as it makes it weaker. Furthermore ‘confidence’ of

100% there is a possibility to be lower to some

Spirometry elements. Although this does not affect

significance and utility of the result. Finally it has

been revealed that FEV1 element has lower

significance on disease recognition as also to

evaluation of it as it has been concerned by authors

in previous section.

5 RECOMMENDATIONS

Initially it has to be clearly understood that the

proposed technique is only to be used as an add-on

tool for better evaluation of health. Even though it

produces an outcome of associations that can

forecast and evaluate patient’s health, other health

examinations have to be taken under consideration.

Additionally the final decision for patient’s health

and treatment must be taken after practitioner’s

overall evaluation. This occurs after combination of

multiple examinations needed as also requirements

that have to be filled.

Through this research both the research questions

that have been set for investigation have been

answered positively. It has been proved that there

are data mining techniques that can find ‘association

rules’ in Spirometry datasets. Moreover it has been

investigated and proved that Apriori algorithm could

be used as most appropriate ‘data mining’ technique

to discover associations among abnormal Spirometry

data. Detailed discussion on how and why this

algorithm is the most suitable for this type of data

are provided in section 3.

Additionally the proposed tool examines and

evaluates patient’s health from Data perspective.

There have to be as many as possible Spirometry

examinations for increased accuracy. A reason for

this is because the algorithms’ process takes into

consideration from 2 to millions of examinations for

analysis. Moreover as it has been discussed

respiratory health is affected by multiple factors

such as weather, daytime, age, weight, etc. With a

small amount of Data the significance may be

reduced but this instance can be overcome by

techniques for small data samples such as Kernel’s

methods and Support Vector Machine (SVM).

Additionally in regular periods, when a

significant amount of Spirometry examinations are

gathered, process of comparison of Data from

different patients has to be carried out. This is

because factors such as climate, global temperature

and multiple factors that affect respiratory and

cannot be measured, change. This could result in

change of associations that provoke severe allergic

hardly controlled Asthma. For avoidance of radical

changes in the association found it has to be checked

in regular periods.

Furthermore because of different environmental

characteristics around globe, technique may have to

be undertaken to samples from different places of

the world. This is for avoiding misleading guidance

from technique caused on environmental changes.

These and some other aspects could be taken into

account and addressed in future work.

Even though this subject it has been approached

by a specific perspective it has been revealed that it

is ‘open’ to be adopted and analysed by different

perspectives.

REFERENCES

Ackan, H., Astashyn, A. and Bronnimann, H., 2008.

Deterministic algorithms for sampling count data. In

Data & Knowledge Engineering, Volume 64, pp 405–

418.

Amato, F., Fasolino, A. R., Mazzeo, A., Moscato, V.,

Picariello, A., Romano, S., and Tramontana, P., 2011.

ConnectionsofReducedPerformanceHealthDataforSeverePersistentUncontrolledAllergicAsthmaTreatedby

Omazulimab

285

Ensuring semantic interoperability for e-health

applications, In Complex, Intelligent and Software

Intensive Systems (CISIS), International Conference

on IEEE, ISBN: 978-1-61284-709-2, June 30 - July 2,

Seoul, Korea, pp 315-320.

Bousquet, J., Rabe, K., Humbert, M., Chung, K.F., Berger,

W., Fox, H., Ayre, G., Chen, H., Thomas, K., Blogg,

M. and Holgate, S., .2007. Predicting and evaluating

response to omalizumab in patients with severe

allergic asthma. In Respiratory Medicine, Volume

101, pp 1483-1492.

Chen, C., Horng, S.J. and Huang, C. P., 2011. Locality

sensitive hashing for Sampling-based algorithms in

association rule mining. In Expert Systems with

Applications, Volume 38, pp 12388–12397.

Cokpinar, S. and Gundem, T.I., 2012. Positive and

negative association rule mining on XML data streams

in database as a service concept. In Expert Systems

with Applications, Volume 39, pp 7503–7511.

Duneja, E. and Sachan, A. K., 2012. A Survey on

Frequent Itemset Mining with Association Rules. In

Computer Applications, Volume 46, pp 18-24.

Guang-Yuana, L., Dan-Yanga, C. and Jian-Wei, G., 2011.

Association Rules Mining with Multiple Constraints.

In Procedia Engineering, Volume 15, pp 1678 – 1683.

Holgate, S., Buhl, R., Bousquet, J., Smith, N., Panahloo,

Z. and Jimenez, P., 2009. The use of omalizumab in

the treatment of severe allergic asthma: A clinical

experience update. In Respiratory Medicine, Volume

103, pp 1098-1113.

Ke, J., Zhan, Y., Chen, X., and Wang, M., 2013. The

retrieval of motion event by associations of temporal

Frequent Pattern growth. In Future Generation

Computer Systems, Volume 29, pp 442-450.

Korn, S., Thielen, A., Seyfried, S., Taube, C., Kornmann,

O. and Buhl, R., 2009. Omalizumab in patients with

databases. In Computer and Information Science,

Volume 23, pp 1-6.

Lee, Y. C., Hong, T. P. and Lin, W. Y., 2005. Mining

association rules with multiple minimum supports

using maximum constraints. In Approximate

Reasoning, Volume 40, pp 44–54.

Lin, K. C., Lia, I.E. and Chen, Z. S., 2011. An improved

frequent pattern growth method for mining association

rules. In Expert Systems with Applications, Volume

38, pp 5154–5161.

Liu, X., Zhai, K. and Pedrycz, W., 2012. An improved

association rules mining method. In Expert Systems

with Applications, Volume 39, pp 1362–1374.

National Heart, Lung, and Blood Institute, 2007. Expert

Panel Report 3: Guidelines for the Diagnosis and

Management of Asthma. In National Asthma

Education and Prevention Program, 28 August.

Nahar, J., Imama, T., Tickle, K. S. and Chen, Y. P. P.,

2012. Association rule mining to detect factors which

contribute to heart disease in males and females

Expert. In Systems with Applications, in press.

Nowak, D., 2006. Management of asthma with anti-

immunoglobulin E: A review of clinical trials of

omalizumab. In Respiratory Medicine, Volume 100,

pp 1907-1917.

Slavin, R., Ferioli, C., Tannenbaum, S., Martin, C., Blogg,

M. and Lowe, P., 2009. Asthma symptom re-

emergence after omalizumab withdrawal correlates

well with increasing IgE and decreasing

pharmacokinetic concentrations. In Allergy and

Clinical Immunology, Volume 123, pp 107-113.

Stout, J. W., Smith, K., Zhou, C., Solomon, C., Dozor, A.

J., Garrison, M. M. and Mangione-Smith, R., 2012.

Learning from a Distance: Effectiveness of Online

Spirometry Training in Improving Asthma Care. In

Academic Pediatrics, Volume 12, pp 88-95.

Tzortzaki, E., Georgiou, A., Kampas, D., Lemessios, M.,

Markatos, M., Adamidi, T., Samara, K., Skoula, G.,

Damianaki, A., Schiza, S., Tzanakis, N., and Siafakas,

N., 2012. Long-term omalizumab treatment in severe

allergic asthma: The South-Eastern Mediterranean

“real-life” experience. In Pulmonary Pharmacology &

Therapeutics, Volume 25, pp 77-82.

Umarani, V. and Punithavalli, M., 2010. Sampling based

Association Rules Mining- A Recent Overview. In

Computer Science and Engineering, Volume 2, pp

314-318.

Umarani, V. and Punithavalli, M., 2011. An Empirical

Analysis over the Four Different Methods of

Progressive Sampling-Based Association Rule

Mining. In Scientific Research, Volume 66, pp 620-

630.

Ykhlef, M., 2011. A Quantum Swarm Evolutionary

Algorithm for mining association rules in large.

Yu, K.M., Zhou, J., Hong,T.P. and Zhou, J.L., 2010. A

load-balanced distributed parallel mining algorithm. In

Expert Systems with Applications, Volume 37, pp

2459–2464.

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

286