Fuzzy Rule-Based Expert System for Assessment Severity of Asthma
- First Online:
- Cite this article as:
- Zolnoori, M., Zarandi, M.H.F., Moin, M. et al. J Med Syst (2012) 36: 1707. doi:10.1007/s10916-010-9631-8
- 249 Views
Prescription medicine for asthma at primary stages is based on asthma severity level. Despite major progress in discovering various variables affecting asthma severity levels, disregarding some of these variables by physicians, variables’ inherent uncertainty, and assigning patients to limited categories of decision making are the major causes of underestimating asthma severity, and as a result low quality of life in asthmatic patients. In this paper, we provide a solution of intelligence fuzzy system for this problem. Inputs of this system are organized in five modules of respiratory symptoms, bronchial obstruction, asthma instability, quality of life, and asthma severity. Output of this system is degree of asthma severity in score (0–10). Evaluating performance of this system by 28 asthmatic patients reinforces that the system’s results not only correspond with evaluations of physicians, but represent the slight differences of asthmatic patients placed in specific category introduced by guidelines.
KeywordsAsthma SeverityAssessmentFuzzyExpert system
Asthma severity is defined as pathology of asthma disease which makes the base for prescribing medicine at primary stage. Asthma control and asthma severity, although closely related in variables applied in determining levels of asthma, but they are distinct concepts . For example a patient with moderate- to-severe asthma could have controlled asthma, whereas a patient with relatively mild asthma, whose therapy is not sufficient, may have relatively poor control of symptoms . Despite improvement in quality of the safe medicine and enhancement knowledge of asthmatic patients, they usually do not get a satisfied improvement at the primary stage of their treatment and unfortunately look for new therapy .
Although epidemiological studies express risk factors associated with asthma severity , the main reasons pertain to underestimating of asthma and underused of sufficient treatment [3–5, 43, 49]. This underestimating usually has unfavorable effects on patients’ quality of life, and growth of lung function . It may finally result in airway remodeling and asthma morbidity [6, 38, 39].
Accurate assessment of asthma severity level requires measuring of objective variables (lung function tests, tests of asthma instability, and skin prick test) and subjective variables (history of respiratory symptoms and quality of life) simultaneously. Although considering all subjective and objective variables presented in the wide range of guidelines brings more accurate result, the variables usually deal with uncertainty that decrease the accuracy of assessment. For example the result of laboratory tests as an objective variable deal with difficulties such as incompleteness, inaccuracy, and inconsistency . In addition, subjective variables such as frequency of nocturnal symptoms usually address patients’ history. These variables are frequently represented as linguistic by patients, so they deal with inherent uncertainty when they are involved in decision making. As a result, considering subjective and objective variables and representing inherent uncertainty enhance reliability of asthma severity evaluation [25–28].
This paper presents an intelligence fuzzy system which provides an appropriate solution for problems of evaluating asthma severity. Knowledge-base of this fuzzy system is composed based on knowledge extracted from a wide range of guidelines covering 10 scores. These scores include Aas score , Hargreave Score , Disease severity score , Artificial intelligence score , Gina score (2006) , Gina score (2007) , Gina score (2008) , Canadian Asthma Consensus Report Score (CACRS) , Prediction of Control State of Asthma Score (PCSAS) , and Expert Panel Report 2 Score (EPRS) .
Output of this system in contrast with output of guidelines is not the assignment of a patient to a specific category. In this system, degree of asthma severity is determined in gradation (0–10) representing the asthma severity’s level from mild-intermittent to severe-persistent. It is obvious that assigning a degree to every patient in this interval represents asthma severity more intelligible for patients.
The rest of this paper is organized as follows: The methodology is presented in Section Methodology, Section Evaluation of system performance contains evaluation system performance, and discussion is discussed in Section Discussion.
Methodology of this system includes phases of knowledge acquisitions, knowledge representation, and evaluation system performance.
System identification of asthma severity evaluation is accomplished based on direct approach [22, 36, 37]. The phase of knowledge acquisition consists of five parts include recognize, collect, interpret, analysis, and design. Output of this stage is general inference network. Two groups of system’s users (asthmatic patients and asthma physicians) are participated in this phase. Physicians perceive the disease from the low level and consider details of asthma mechanism and pathology, but patients view it from high level concerning on major consequences such as pain in chest .
As a result considering both views are very important for building knowledgebase. Latent sources such as books, papers, and medical websites are used as complement sources.
Regard to the output of this stage, important variables are identified and classified in 5 modules. Variables of frequency of daily symptoms, frequency of nocturnal symptoms, and Dyspnoea in exercise are categorized in module of respiratory symptoms severity (RSS), Forced Expiratory Volume in 1 s (FEV1), Forced Vital Capacity (FVC), and FEV1/FVC are categorized in module of bronchial obstruction (BO), Peak Expiratory Follow (PEF), and Frequency of Severe Exacerbations are placed in module of asthma instability (AI), Frequency of Missing days of School/work, Frequency of Missing days of daily Activities, and Frequency of looking for new treatment are placed in the class of quality of life (QL). Variables of skin prick test, IgE value, response to short B2-agonist, and modules of RSS, BO, AI, and QL are classified in asthma severity module.
- General inference network: General Inference Networks (GIN) provides a graphical representation of inputs and outputs drawn as nodes. Figure 2 depicts the GIN of variables and classes (modules) indentified in the phase of knowledge acquisition.
Production rules: various combination of inputs and output in GIN (Fig. 2) are formed the production rules of this system. These rules have been generated using interview techniques, task performance and protocols, and questionnaires and surveys. Examples of rules related to every module are represented in the Appendix.
Representing uncertainty: Knowledge acquisition (intuition and judgment), Knowledge representation, and incomplete information are sources of uncertainty in representing asthma severity knowledge. Fuzzy logic is used in this system to answer three questions: 1) how to present linguistic and numeric values? 2) How to combine two or more piece of uncertain data?, and 3) how to design inference engine using uncertain data? . Identification of fuzzy sets of uncertain variables are calculated by asking the physicians for determining the points with (MF = 0) and points with (MF = 1). These two points are connected together making triangular or trapezoidal membership functions that are common in representing fuzzy sets because of their computational efficiency .
System inference engine: Mamdani inference engine  has been used for parallel processing using maximum and minimum operation. The mechanism of centriod is used for deffuzification of fuzzy output sets.
Module of Respiratory Symptoms Severity (RSS)Estimating severity of respiratory symptoms is important for evaluating asthma severity [7, 8, 12–14, 16, 30]. Variables of frequency of daily and nocturnal symptoms (including cough, wheeze, tightness of chest) during 1 month, and status of shortness of breath (dyspnoea) during activities or at rest are represented as the rules’ antecedents of this module. Output of this module, degree of respiratory symptoms, is represented as linguistic values from normal to more in gradation of (0–10). Considering history of respiratory symptoms usually contains uncertainty. As a result, it doesn’t present acceptable reliability for evaluating asthma severity [23, 42]. Representing these variables using fuzzy techniques could improve the process of evaluation. Table 1 shows variables of rules’ antecedents and rules’ consequents of RSS with Linguistic values and fuzzy intervals related to these variables.Table 1
Rules’ antecedents and rules’ consequents of respiratory symptoms module
FDS = Frequency of Daily Symptoms
N = Never
Se = Seldom
So = Sometimes
O = Often
F = Frequently
A = Always
FNS = Frequency of Nocturnal Symptoms
N = Never
Se = Seldom
So = Sometimes
O = Often
F = Frequently
A = Always
DE = Dyspnoea in Exercise
Dy1 = can walk indefinitely(very_low)
Dy2 = gets shortness of breath with strenuous exercise(low)
Dy3 = shortness of breath with moderate exercise such as climbing one or two flight of stairs (Medium_low)
Dy4 = shortness of breath with minimal exercise such as climbing one half to one flight of stairs (Meduium)
Dy5 = shortness of breath with personal work (such as dressing) (Medium_high)
Dy6 = shortness of breath at rest (high)
Degree of respiratory symptoms severity
N = Normal
MI = Mild
MIMO = Mild_to_Moderate
MO = Moderate
MOMO = Moderate_to_More
M = More
Module of Bronchial Obstructions (BO)Variables of Forced Expiratory Volume in 1 s (FEV1), Forced Vital Capacity (FVC), and FEV1/FVC are considered as indicators of spirometry test for measuring of airflow obstruction [7, 8, 11, 13–16, 31]. Results of these tests may deal with some uncertainties [16, 24–28, 36] that have negative effects on accurate assessment of degree of bronchial obstruction and as a result assessment of asthma severity level. Table 2 presents variables of FEV1, FVC, and FEV1/FVC as rules’ antecedents with related fuzzy intervals. Figure 4 presents some of the rules, and membership functions of variables depicted in this module. Output of this module is level of bronchial obstruction.Table 2
Rules antecedents and rules consequents of bronchial obstruction module
Bronchial obstruction variables
Degree of bronchial obstruction
Module of Asthma instabilityResults of Peak Expiratory Flow (PEF) during 2 weeks and frequency of Severe Exacerbation (FSE) leading to hospitalization have been recognized as power tools for determining asthma instability [7, 8, 11, 13–16, 29]. A patient with even one time of asthma exacerbation during a year has problem of asthma instability. Table 3 represents input variables, (rule’s antecedents) and degree of asthma instability as output variable (rule’s consequence) with their related categories and fuzzy intervals. Values of these variables are entered by patients or physicians based on the results of PEF test and frequency of severe asthma exacerbations during a certain interval (1 month). Figure 5 presents some of the rules, and membership functions of variables depicted in this module.Table 3
Rules’ antecedents and rules’ consequents of asthma instability module
Asthma instability variables
FSE = Frequency of Severe Exacerbation
Degree of asthma instability
Module of Quality of Life (QL)In a narrow sense, accurate evaluation of asthma severity requires taking into account how asthma limits activities . In broad sense considering asthmatic patient’s view of life is very important for measuring asthma severity [32, 41]. Disregarding this variable can decrease the accurateness of severity assessment [1, 40].To evaluates the quality of life in asthmatic patients, following variables are considered:
FMS/FMW: Frequency of Missing days of School/Work because of asthma
FMA: Frequency of Missing any daily Activities (such as playing, going to a friend’s house, or any interesting activities)
FLT: Frequency of Looking for new Treatment
Table 4 contains variables of FMS/FMW, FMA, and FLT as rules’ antecedents and degree of quality of life as rules’ consequents of this module. Figure 6 presents some of the rules, and membership functions of variables depicted in this module.
Rules’ antecedents and rules’ consequents of quality of life
Quality of life’s variables
FMS/FMW = frequency of Missing days of School/Work
FMA = frequency of Missing days of daily Activities
FLT = Frequency of looking for new Treatment
One = <2
More = >1
Degree of quality of life
Module of asthma severity
Module of asthma severity is responsible for aggregation of outputs received from sub-modules including RSS, BO, AI, QL and variables of IgE, Skin Prick Test (SPT), and Response to Medicine (RM).
Severity of IgE [20, 45], and Skin prick test are associated with severity rate of atopy, and as a result, level of asthma severity. Outputs of these tests usually are presented by linguistic values of low, medium, and high that are associated with imperfection and imprecision.
Response to medicine is correlated with severity of asthma. Asthmatic patients with high severity usually have poor reaction to inhaled B2-agonist . This variable is usually represented with linguistic values as poor and fair involving uncertainty for decision making in assessment asthma severity.Table 5 represents uncertainty for these variables. Figure 7a,b and c depict the membership function of linguistic values related to IgE, Skin Prick Test (SPT), and response to medicine.Table 5
Linguistic values related to IgE, skin prick test, and response to medicine
Linguistic value related to SPT & IgE,and RD
SPT = skin prick test
L = Low
ML = Medium_low
M = Medium
MH = Medium_high
H = High
L = Low
ML = Medium_low
M = Medium
MH = Medium_high
H = High
RM = response to Medicine
F = Fair
RP = Rather Poor
P = Poor
These variables, in addition to output of the modules make input variables of assessment of asthma severity. Aggregation of these input fuzzy sets makes output fuzzy set of this system.Defuzzification of this fuzzy set is a real number in the interval (0–10) representing degree of asthma severity. This interval is classified in five categories (introduced by guidelines) to present understandable assessment for patients and physicians and provides criteria for evaluating performance of this system in comparison with physicians’ evaluation. These categories and their intervals are shown in Table 6. These intervals have been determined based on the opinion of asthma physicians in asthma, immunology, and allergy research center of Tehran, Iran University.Table 6
linguistic categories of asthma severity and related intervals
MI = Mild Intermittent
MP = Mild Persistent
MOP = Moderate Persistent
SP = Severe Persistent
Evaluation of system performance
To compare system’s results and physician’ evaluations of severity, system outputs are classified in four categories (Table 6). Kappa coefficient (k)  is calculated for comparison between categories of system output and physician’s evaluation, the obtained results show the value of k = 1, reinforcing the complete corresponding between system’s evaluation and physician’s evaluation based on the defined intervals in Table 6.
Fuzzy expert system presented in this paper considers problems of underestimating of asthma severity that result in wrong prescription of medicine and finally uncontrolled asthma. Uncertainty of variables used for estimating of severity, disregarding of some main variables, and assigning patients to limited categories (3–5) for prescription cause this problem.
Developed intelligence systems such as assessment severity of asthma and identification of triggers factors , and expert system for assessment severity based on introducing new score  usually translate the guidelines’ knowledge to computer systems while the mentioned problems still persist.
Representing uncertainty by fuzzy technique, parallel processing of variables elicited from different guidelines and providing a gradation (0–10) as system output that assigns a specific grade to every patient are features of this system that address difficulties of evaluation of asthma severity. Because of inexistence of comprehensive medical records that include all variables introduced in this paper, sample of patients’ data for test of this system is restricted to 28 asthmatic patients referring to asthma, immunology, and allergy center (Emam khomaini hospital, Tehran, Iran) and have been examined by asthma physicians. Comparison physicians’ evaluations with system’s results (for these 28 patients) reinforces he acceptable performance of this system. In this comparison, system outputs have been classified in 4 categories regard to intervals determined by asthma physicians. These categories which are introduced by guidelines include Mild Intermittent, Mild Persistent, Moderate Persistent, and Severe persistent. Kappa coefficient calculated for this comparison shows value of 1 presenting the thorough corresponding between physicians’ evaluation and system’s evaluation.
The knowledge base of this system has been designed as modular. Every module can be used independently for evaluation of indicators including bronchial obstruction, severity respiratory symptoms, asthma instability, and quality of life. This characteristic facilitates the process of system maintenance.
This system can be used without requiring of lab data (lung function test, and atopy test), so it can be applied in primary care setting for assessment severity of asthma by general physician. However the result of system without laboratory data may not have the precision when all variables have been considered for evaluating severity. Future work can evaluate performance of this system by increase of sample size. In order to improve system accuracy in primary care setting (without laboratory data), the concentration of representing variables can be restricted to symptoms data and historical data.
I would like to appreciate all who gave me the possibility to complete this paper. I want to thank the physicians and staff of Immunology, Asthma & Allergy Research center for their collaboration in providing necessary knowledge. I thank Dr Heydarnejad and Dr. Fazlollahi for their guidance, Mehdi Taherian for contribution in knowledge representation I appreciate Zahra Zolnoor, Ali Zolnoor, and Mohammad Reza Zolnoor for providing general information. Finally I thank Neda Kharghani for reviewing this paper and providing useful comments.