Clinical Decision Support System in Virtual Clinic

Transcription

1 Clinical Decision Support System in Virtual Clinic Atta-ur-Rahman Department of Computer Science, College of Computer Science and Information Technology, Imam Abdulrahman Bin Faisal University, P.O. Box 1982, Dammam, Kingdom of Saudi Arabia. Abstract: Healthcare facilities in Pakistan are not sufficient, particularly in rural or remote areas. These areas show poorer health outcomes than urban areas due to lack of trained physicians, clinics and healthcare services. This is mainly because, most of the healthcare practitioners don t like to practice in these areas. In contrast, a huge number of lady doctors don t practice after their graduation due to various reasons, thus major portion of medical human resource is not utilized. Moreover, the information and communication technologies (ICT) in Pakistan are becoming popular and affordable day by day. Many developing countries like Pakistan have started the implementation of telemedicine projects. This research work focuses on connecting the remote areas of Pakistan to the urban areas practices and doctors especially the non-practicing lady doctors through ICT by establishing virtual clinics in remote areas. In addition to that, the virtual clinic decision support system (VCDSS) facilitates the users in two ways. Firstly, automatically assigning the appropriate most doctor for patient based on his ranking, experience and specialization etc. using a fuzzy rule based system (FRBS). Secondly, it helps the registered doctors at the time of prescribing a patient by suggesting the medicines according to the patient s symptoms and history. It is done by using the knowledge-base of past diagnoses/prescriptions by the same doctor as well as other doctors and British National Formulary (BNF) using Apriori Algorithm and Inductive Learning Algorithm. The results are shown using sample dataset. Keyword: CDSS, Virtual Clinic, Apriori Algorithm, Data mining, WEKA 1. Introduction In Pakistan, the healthcare facilities are insufficient in rural areas while the population of rural areas is more than sixty percent. Patients face huge inconvenience in travelling a long way to the urban areas for treatment especially with poor transportation facilities. In this situation, information and communication technology (ICT) can help to extend better healthcare facilities in the rural areas. Any medical activity involving a factor of distance is defined as telemedicine. Stanberry (2001) [1] proposed that the fast and advance development of ICT has changed the world. Telematics applications such as sending images, telephone call is removing the barrier of distance but it will lead to the loss of the traditional healthcare framework and the comfort one can only have by face to face consultation. Most successful examples of educating medicine at a distance, includes the Mayo clinic, an educational program that was delivered by satellite2. Sood et al (2007) [2] proposed that telemedicine can be defined in different ways by researchers and technology users. Modern telemedicine is a branch of e-health that transfers healthcare services from one location to another by ICT. Many institutions around the world are moving out research activities in the field of telemedicine including Brunel University, University College Cork (Ireland), University of Miami, University of Virginia, Harvard and Boston University, Imperial College London, Harvard and Boston University, the Johns Hopkins University, University of Texas etc. Armstrong and Haston [3] presented an approach for telemedicine technology. They have proposed a telemedicine project in which connection was setup between the two departments of a remote and urban hospital. Videoconferencing, teleradiology and telepresence were the key features of the link. Connection was based upon ISDN and satellite link. Within one year nearly 120 teleconsultations took place between the hospitals. Results indicated that tele-consultation had enhanced patient care. Estimated cost saving was about sixty-five thousand USD. Carson (1998) [4] described the role of 2018 by the author(s). Distributed under a Creative Commons CC BY license.

2 telemedicine in the management of diabetes and chronic diseases. It may include the evaluation of decision support systems (DSS), methodology and framework by using system modelling as a key approach for designing system infrastructure. This paper reviews current evidence and provides clinical applications where telemedicine is beneficial and cost-effective. In [5] authors proposed a wireless telemedicine system named as WISTA offering low cost and easy to deploy systematic framework. It will assist patients during unfavourable conditions. Multimedia consultation occurs between multiple disasters sites to the control centre of the disaster. Proposed system supports real-time consultation offering quick treatment procedures. It uses a hierarchical architecture for implementation. For thick disaster areas, a layered structure is functional to support information distribution from large number of patients. Moreover, the performance of the system is computed using OPNET simulation. Other than clinical decision support systems, there exist DSS for other domains [6-8] as well. 1.1 Telemedicine Projects in Pakistan Authors in [9] has proposed telemedicine approach in the rural areas of Pakistan. He proposed to connect these rural areas with these non-practicing lady doctors through any information and communication technology whether it is internet, GSM, WiMAX, Satellite etc. For this sake, virtual clinics will be established in the remote and ruler sites, patients will visit these clinics a nursing staff would be available to take all necessary information that will be sent by a smart phone to a registered doctor through a central system. Doctor in return would prescribe and prescription will be sent back by smartphone to virtual clinic via same central system. Memon et al (2004) [10] has proposed a telemedicine system that support both store and forward and real time consultation methods for quick and cost effective response. Relevant patient information has been sent using ISDN and satellite links. The system approaches towards the mobile doctor in all decisions. The system has various advantages like reduces the cost and time, increases. Pakistan is a thickly populated country where large population resides in rural areas. These areas lack basic health services and specialists. Telemedicine projects in developing countries like Pakistan are introducing new ways of improved health care services. Malik (2007) [11] presents an approach for information and communication. It highlights the projects established by Pak-US collaboration, SUPARCO a satellite based telemedicine centres and establishing global telemedicine network which connect all medical college hospitals and district hospitals. Karim et al (2011) [12] proposed a virtual medicine approach that is composed of a clinical decision support system deployed at rural area, to provide medical facilities in an active way at reduced cost. This CDSS diagnosis the patient and suggest prescription as well. The cases where it is unable to prescribe, it sends to the doctor and store response in its database for future use. In this research paper, complete setup of virtual telemedicine approach has been discussed. From the discussion, it is apparent that a comprehensive, practical and cost-effective telemedicine solution equipped with clinical decision support system (CDSS) is still a need of remote areas of Pakistan. This is addressed in this paper. Rest of the paper is organized as follows. Section 2 highlights the proposed system model, section 3 describes the research methodology used to carry out the research, section 4 describes the proposed approach while section 5 concludes the paper. 2. System Model For collecting patients information from rural communities, the system consists of clinic side interface. The patients from rural areas to the virtual clinic where the health works available in clinic to collect the patient

3 basic data if the patient is new then registered to the system if patient profile already exists then open the patient s profile. Collect the symptoms and vitals of the patients and submit it to the system. Proposed system finds doctor/physician for the patient and sends the data to the doctor along with the suggested medications and investigations. The submitted data is also stored in our history data base for future use. The doctor side interface is for registered doctors where he/she can view their incoming patients along with the suggestions by the system. He may then modify the suggested investigations and submit the treatment and to the system. The system stored the information in the database as well as sends it back to the virtual clinic. Virtual clinic staff can collect the suggested treatment and prescription by the doctor and pass it to the patients. VC-DSS system suggests the doctor based on the doctor availability and ranking. Ranking of registered doctors is done by our system on the base of their experience, specialization, relevance and quick response time of the history. The system will store all the information of the registered patients, doctors, investigations and prescription for future use of system like to maintain patients and doctors history, doctor ranking etc. Abstract architecture is shown in the following fig-1. Figure 1: Architecture of Victual Clinic with DSS The VC-DSS Server is the backbone of whole Decision Support System in Virtual Clinic. This is a main web server which receives patient s information from the virtual clinic passed by health workers serving in virtual clinics in rural areas. The VC-DSS checks the patient s complete information received from virtual clinic for errors and abnormalities and passed it to decision support system as well as store the information in the patient s database. After DSS analysis and processing it is finally delivered to the appropriate doctors for consultation and prescription with suggested medicine and advices. The treatment and prescription given by consultant is then forwarded to virtual clinic side as well as store in the database. DSS is a decision support tool that decides that the patient is sends to which doctor based on the information, investigation, symptoms of patients and doctor ranking, efficiency and successful recommendation. It also helps for consultants in a way that along with patients it also suggests the medication and treatment. In this system DSS have the following functions.

4 Assign the doctor Ranking of doctor Suggest the medications to the doctors 3. Research Methodology 3.1 Apriori Algorithm Apriori algorithm is used for the problem of association rule mining, the classic algorithm. In the proposed system, it is used to assign consultant to the patients, suggest medications using three types of suggestions based on system database history (from all the doctors), doctor s own history and using British National Formulary (BNF). Apriori uses association rule having a support level and a confidence level. Support is the percentage of the population which satisfies the rule. If the percentage of the population in which the attendant is satisfied, then the confidence is that percentage in which the consequent is also satisfied. 3.2 Inductive Learning Algorithm Inductive Learning algorithm (ILA) is also investigated to assign consultant to the patients, suggest medications using three types of suggestions based on system database history, doctor s own history and using British National Formulary (BNF). In ILA sub tree for each class is established. Rules can be generated from these sub trees. Once the association is selected for rule, it does not consider for future rules. If all are unmarked, go for the combination in sub tree, select maximum group of combination. 3.3 Assigning The Consultant Decision support system goal is to perform analysis, make proper decision and quick response. DSS used an efficient algorithm decide the new patient is refer to which consultant. The algorithm used the patient information, investigation, vitals, decease type, and previous history of the database as well as the doctors ranking and specialty, to refer to the appropriate consultant. DSS also maintain the history of diagnostics, treatment, medication, advice and prescription of the doctors as well as their response time and effectiveness of prescription and medication feedback for future use i.e. patients history and ranking of doctors maintained. DSS monitors the patients and doctor s important data to make the better decision in future. Apriori algorithm and Inductive Learning algorithm are used to establish the rules to assign doctors to the patients by using prior knowledge and history from database systems. Patient s important data for example patient s basic information, history, blood pressure, temperature, symptoms list, images, reports history, diabetes and health situation etc. The doctors important data include the ranking, specialty, diagnostics, prescription, and effectiveness of the treatment and medication. Ranking of registered doctors is one key role of the DSS system. 3.4 Proposed FRBS assisted Ranking The ranks assign to the doctors are excellent, good, average, low, very low based on various factors like their response time, effectiveness of the treatment and experience. When a patient visits VC, DSS dynamically allocates the doctor with the highest rank depending on his daily limit. In case the limit is over then second

5 highest rank holder is approached etc. This helps in control the workload and increases the system efficiency by using knowledge driven decision-making system. In this regard, a fuzzy rule based system (FRBS) for efficient ranking calculation is proposed. The FRBS is two folds. First the effectiveness of the doctor s previous prescriptions is calculated that depends on the patients feedback and super-doctor s level of satisfaction towards the doctor s prescription for a specific case. Once the effectiveness is calculated, two further factors are involved to calculate the final rank. Those factors are doctor s experience and response time. Fig-x shows the model of the proposed FRBS. Figure 2: FRBS conceptual model Components of the FRBS In first level of FRBS, the fuzzy there are two input variables namely patient-feedback and super-doctor-review, which results in fuzzy output variable effectiveness. In turn, the output of FRBS-I (effectiveness), along with experience and response-time become inputs to FRBS-II, which eventually generates the fuzzy output variable rank. Brief description of components used in both FRBS is enlisted in table-1. Table 1: Parameters used in FRBS # Parameter Detail 1 Number of input variable and membership 2 (3,3) functions in FRBS1 2 Number of output variable and membership 1 (5) functions in FRBS1 3 Number of rules in FRBS1 9 4 Number of input variable and membership 3 (3,3,5) functions in FRBS2 5 Number of output variable and membership 1 (7) functions in FRBS2 6 Number of rules in FRBS Type of fuzzifier Triangular 8 Type of Fuzzy Inference Engine used Mamdani 9 Type of de-fuzzifier Centre average de-fuzzifier 10 Rule generation principle Heuristic Fig-3 (a) and (b) show the input variables of FRBS-1 and fig-3 (c) shows the output variable of FRBS-1. Similarly, fig-3 (d) and (e) show the input variables of FRBS-2 and fig-3 (f) shows the final output of the both FRBS.

6 (a) Input variable Patient Feedback (b) Input variable Super-doctor reivew (c) Output variable Effectiveness which is also input to FRBS2 (d) Input variable Response (e) Input variable Experience (f) Output variable Rank Figure 3: Input and output variables used in both FRBS Fig-4 shows the rule surfaces of FRBS-1 and FRBS-2. For instance, fig-4 (a) shows rule surface for FRBS-1 which shows that Effectiveness is directly proportional to patient-feedback and super-doctor-review. Moreover, both have equal impact on output variable (effectiveness). Fig-4 (b) shows the impact of effectiveness and response on the rank of the doctor. In this regard, both have impact on rank but effectiveness has more impact compared to response time. Similarly, fig-4 (c) and (d) shows the joint impacts of experience-response and effectiveness-experience on the rank. From this we can see that experience and response have almost same

7 effect while in effectiveness-experience case, effectiveness is given more weightage compared to the experience on the rank of doctor, respectively. (a) Rule surface of FRBS-1 (b) Rule surface of FRBS-2 for Effectiveness and Response (3) Rule surface of FRBS-2 for Experience and Response (4) Rule surface of FRBS-2 for Effectiveness and Experience Figure 4: Rule surfaces of both FRBS 3.5 Proposed CDSS One of the facilities provided by this system is to suggest the medicine to the doctors while he is prescribing the patients. This system provides three suggestions to the doctors by using the overall history in the system, the doctor s own history and the BNF database. Apriori and Inductive Learning Algorithms are investigated in this regard. The doctor can use any of the suggestions provided by the system or can modify an existing suggestion based on current symptoms of the patient and effectiveness of the medication. The modifications will also be stored in the system for future use as a new instance. 4. Results and Discussion A sample data-set given below and the DSS results are given below by using Apriori algorithm and Inductive Learning Algorithm. Apriori algorithm is the mining tool to find the frequent itemsets for association rules. This algorithm uses the prior knowledge to find the frequent itemsets to establish the rules. Here a sample dataset is used to establish the rules examples. This dataset is given in table-2.

8 Table 2: Sample Dataset from Database Dr. Pt1 M 55 Asthma Shortness of breath, chest pain pt2 M 35 Diabetes Hunger, increased urination, thirst Pt3 M 60 Heart Attack Heartburn, Arm pain, Back pain Pt4 M 21 Dengue Fever, Headache, Muscle Pain Pt 5 F 18 Pneumonia Cough, Shortness of Breath, Fever Theophylline, Chlorpropamide Aspirin Acetaminophen Amoxicillin, Augmentin A B D E F Pt 5 M 3 Dengue Muscle pain, Fever Tylenol G Pt7 M 5 Pneumonia Fever, chest pain Augmentin G Pt8 M 35 Diabetes Hunger, Thirst Chlorpropamide B Pt9 F 26 Dengue Headache, Muscle pain Acetaminophen E Pt10 F 2 Dengue Headache, Muscle pain Tylenol G PT1 M 1 Pneumonia Fever, Chest Pain Augmentin G Pt12 M 19 Pneumonia Fever, Chest Pain Amoxicillin, Augmentin F Pt13 M 4 Pneumonia Chest Pain Augmentin G Pt14 M 15 Diabetes Hunger, Thirst Chlorpropamide B Pt15 F 18 Pneumonia Fever, Chest pain Amoxicillin F 4.1 Suggest the Doctor Using Apriori Algorithm Min Confidence=50%

9 Min Support=20% Age grouping Rules Examples 0~10 => 1 10~20=> 2 20~30=> 3 30~40=> 4 40~50=> 5 50~60=> 6 >60=> 7 Rules (1, Pneumonia => Dr. G) =(3/15)/(5/15)= % Confidence Rules (1=> Dr. G) =(5/15)/(5/15)= 1 100% Confidence Rules (Heart Attack =>Dr. D) = (1/15)/(1/15)= 1 100% Confidence Rules (Diabetes => Dr. B) = (3/15)/(3/15)= 1 100% Confidence Using Inductive Learning Algorithm Dr. A -Sub tree Table 3: Inductive Learning Subtrees-Assign Doctor Dr. Pt1 M 55 Asthma Shortness of breath, chest pain Theophylline, A Dr. B-Sub tree Dr. pt2 M 35 Diabetes Hunger, increased urination, thirst Chlorpropamide B Pt8 M 35 Diabetes Hunger, Thirst Chlorpropamide B Pt14 M 15 Diabetes Hunger, Thirst Chlorpropamide B Dr. D-Sub tree Dr. Pt3 M 60 Heart Attack Heartburn, Arm pain, Back pain Aspirin D

10 Dr. E-Sub tree Dr. Pt4 M 21 Dengue Fever, Headache, Muscle Pain Acetaminophen E Dr. F-Sub tree Dr. Pt 5 F 18 Pneumonia Cough, Shortness of Breath, Fever Pt12 M 19 Pneumonia Fever, Chest Pain Amoxicillin, Augmentin Amoxicillin, Augmentin F F Pt15 F 18 Pneumonia Fever, Chest pain Amoxicillin F Dr. G-Sub tree Dr. Pt 5 M 3 Dengue Muscle pain, Fever Tylenol G Pt7 M 5 Pneumonia Fever, chest pain Augmentin G Pt9 F 6 Dengue Headache, Muscle pain Acetaminophen G Pt10 F 2 Dengue Headache, Muscle pain Tylenol G PT1 M 1 Pneumonia Fever, Chest Pain Augmentin G Pt13 M 4 Pneumonia Chest Pain Augmentin G Rules Generation Rules (Disease=Asthma) => Dr. A Comparison Rules (Disease=Diabetes) => Dr. B Rules (Disease=Heart Attack) => Dr. D Rules (Age<10) => Dr. G Rules (Disease=Dengue and Age> 10) => Dr. E Rules (Disease=Pneumonia and Age> 10) => Dr. F Table-3 shows the ILA subtrees against each doctor. Table-4 provides the comparison between the both schemes. The common suggestions got the highest priority (0) and appear top of the list, while the individual suggestion got the medium priority (0.5). In case where, there is no suggestion for a scenario, which will be

11 treated as lowest priority (1) and this will appear in the lower part of the list. Comparison shows that ILA performs better in terms of number of rules. Table 4: Comparison Rule ILA Apriori Algorithm Priority Disease=Asthma A 0.5 Disease=Diabetes B B 0 Disease=Heart Attack D D 0 Age<10 G G 0 Disease=Dengue and Age> 10 E 0.5 Disease=Pneumonia and Age> 10 F 0.5 Disease=Pneumonia and Age< 10 G Suggest Medications by Local History Using Apriori Algorithm Min Confidence=50% Min Support=20% Age grouping 0~10 => 1 10~20=> 2 20~30=> 3 30~40=> 4 40~50=> 5 50~60=> 6 >60=> 7 Rules Examples Rules (M,1, Chest Pain => Pneumonia) =(3/15)/(6/15)= % Chance of Pneumonia Rules (M,1, Chest Pain => Augmentin) =(3/15)/(5/15)= % Chance of Augmentin Rules (M,1, Chest Pain => G) = (3/15)/(5/12)= % Chance of Dr. G Rules (2, Fever => Pneumonia) = (3/15)/(5/15)= % Chance of Pneumonia

12 4.2.2 Using Inductive Learning Algorithm Table-5 shows the ILA sub trees to suggest medication from the local history. Theophylline - Sub tree Table 5: Inductive Learning Sub Trees-Suggest Medicine from Local History Pt1 M 55 Asthma Shortness of breath, chest pain Theophylline Chlorpropamide - Sub tree pt2 M 35 Diabetes Hunger, increased urination, thirst Chlorpropamide Pt8 M 35 Diabetes Hunger,Thirst Chlorpropamide Pt14 M 15 Diabetes Hunger,Thirst Chlorpropamide Aspirin- Sub tree Pt3 M 60 Heart Attack Heartburn,Arm pain,back pain Aspirin Acetaminophen Sub tree Pt4 M 21 Dengue Fever, Headache, Muscular Pain Acetaminophen Pt9 F 26 Dengue Headache, Muscle pain Acetaminophen Amoxicillin - Sub tree Pt 5 F 18 Pneumonia Cough, Shortness of Breath, Fever Amoxicillin, Augmentin Pt12 M 19 Pneumonia Shortness of Breath, Fever Amoxicillin, Augmentin Pt15 F 18 Pneumonia Shortness of Breath, Fever Amoxicillin

13 Augmentin- Subtree Pt 5 F 18 Pneumonia Cough, Shortness of Breath, Fever Amoxicillin, Augmentin Pt12 M 19 Pneumonia Shortness of Breath, Fever Amoxicillin, Augmentin Pt7 M 5 Pneumonia Fever, chest pain Augmentin PT1 M 1 Pneumonia Fever, Chest Pain Augmentin Pt13 M 4 Pneumonia Chest Pain Augmentin Tylenol Subtree Pt 5 M 3 Dengue Muscle pain, Fever Tylenol Pt10 F 2 Dengue Headache, Muscle pain Tylenol Rules Generation Rules (Disease=Asthma) => Theophylline Rules (Disease=Diabetes) => Chlorpropamide Rules (Disease=Heart Attack) => Aspirin Rules (Desease= Dengue and Age<10) => Tylenol Rules (Disease=Dengue and Age> 10) Acetaminophen Rules (Disease=Pneumonia and Symptom=Shortness of Breath) => Amoxicillin Rules (Disease=Pneumonia and Symptom=Chest Pain) => Augmentin Comparison Same behavior can be observed here, ILA performs better in terms of number of rules due to high number of sub-trees that are based on different medicines. Still those suggestions will appear on top of the list in the doctor graphical user interface (GUI) where both algorithms agree and then there comes the suggestions with at least one of the algorithms agrees and eventually the remaining suggestions that are usually due to less number of instances of that type. 4.3 Suggest Medications by Doctors History Table-6 shows sample dataset for doctor s history.

14 Table 6: Data Set Doctors History Pt1 M 5 Flu Fever Arinac pt2 M 3 Loose Motions Motions Navidate Pt3 M 6 Flu Fever Arinac Pt4 M 2 Dengue Fever, Headache, Muscular Pain Tylenol Pt 5 F 8 Pneumonia Chest Pain, Fever Augmentin Pt 5 M 3 Dengue Muscle pain, Fever Tylenol Pt7 M 5 Pneumonia Fever, chest pain Augmentin Pt8 M 5 Loose Motions Motions Navidate Pt9 F 6 Dengue Headache, Muscular pain Tylenol Pt10 F 2 Dengue Headche, Muscle pain Tylenol Rules Examples Using Apriori Algorithm Rules(Dengue = > Tylenol) =(4/10)/(4/10)=1 100 % chance of Tylenol Rules (Chest Pain => Augmentin) =(2/10)/(2/10)= % chance of Augmentin Rules (Pneumonia => Augmentin) =(2/10)/(2/10)= % chance of Augmentin Rules (Fever => Arenac) =(2/10)/(4/10)=.5 50 % chance of Arenac Using Inductive Learning Algorithm Table-7 shows the ILA subtrees to suggest medicine from the doctor s history. Table 7: Inductive Learning Subtrees Arenac -Sub tree Pt1 M 5 Flu Fever Arinac Pt3 M 6 Flu Fever Arinac

15 Navidate- Sub tree pt2 M 3 Loose Motions Motions Navidate Pt8 M 5 Loose Motions Motions Navidate Tylenol- Sub tree Fever,Headache,Muscle Pt4 M 2 Dengue Pain Tylenol Pt9 F 6 Dengue Headche, Muscle pain Tylenol Pt 5 M 3 Dengue Muscle pain, Fever Tylenol Pt10 F 2 Dengue Headche, Muscle pain Tylenol Augmentin Sub tree Cough,Shortness of Pt 5 F 8 Pneumonia Breath,Fever Augmentin Pt7 M 5 Pneumonia Fever, chest pain Augmentin Rules Generation Rules (Disease=Flu) => Arenac Rules (Disease=Loose Motions) => Navidate Rules (Disease=Dengue) => Tylenol Rules (Disease=Pneumonia) => Augmentin Table-8 shows the comparison of Apriori Algorithm and Inductive Learning Algorithm in terms of suggesting medicines from doctor s history. Highest priority (0) is given to the common suggestions and lower is assigned to distinct suggestions that is either by Apriori or Inductive Learning Algorithm not both. In case there a disease with no suggestions by both algorithms is assigned a lowest priority. However, they will appear as an instance. Table 8: Comparison Disease ILA Apriori Algorithm Priority Flu Arenac 0.5 Loose Motion Navidate 0.5 Dengue Tylenol Tylenol 0 Pneumonia Augmentin Augmentin 0

16 4.4 Suggest Medications using BNF Dataset Table-9 shows the dataset extracted from British National Formulary (BNF). Table 9: Data Set for BNF Disease Symptoms Medication Diabetes Shortness of breath, chest pain Chlorpropamide Diabetes Hunger, increased urination, thirst Chlorpropamide Heart Attack Heartburn,Arm pain,back pain Aspirin Dengue Fever,Headache,Muscle Pain Acetaminophen Pneumonia Cough,Shortness Of Breath,Fever Amoxicillin, Augmentin Dengue Muscle pain, Fever Tylenol Pneumonia Fever, chest pain Augmentin Diabetes Hunger,Thirst Chlorpropamide Dengue Headche, Muscle pain Acetaminophen Dengue Headche, Muscle pain Tylenol Rules Examples Using Apriori Algorithm Rules for (Dengue, Tylenol) Rules(Tylenol = > Dengue ) =(2/10)/(4/10)= % Chance of Tylenol Rules(Dengue = > Tylenol) =(2/10)/(2/10)=1 100 % Chance of Dengue Rules (M,1, Chest Pain => Augmentin) =(1/10)/(2/10)= Chance of Augmentin Using Inductive Learning Algorithm Table-10 shows the ILA subtrees from the dataset given in table-9. The dataset belongs to BNF databased that contains knowledge about the medications against the diseases already.

17 Table 10: Inductive Learning Sub trees to suggest medicine Using BNF Chlorpropamide- Sub tree Disease Symptoms Medication Diabetes Diabetes Shortness of breath, chest pain Hunger, increased urination, thirst Chlorpropamide Chlorpropamide Diabetes Hunger, Thirst Chlorpropamide Aspirin- Sub tree Disease Symptoms Medication Heart Attack Heartburn, Arm pain, Back pain Aspirin Acetaminophen- Sub tree Disease Symptoms Medication Dengue Headache, Muscle Pain Acetaminophen Dengue Headche, Muscle pain Acetaminophen Augmentin- Sub tree Disease Symptoms Medication Pneumonia Cough, Shortness of Breath, Fever Amoxicillin, Augmentin Pneumonia Fever, chest pain Augmentin Tylenol- Sub tree Disease Symptoms Medication Dengue Muscle pain, Fever Tylenol Dengue Fever, Muscle pain Tylenol Amoxicillin Sub tree Disease Symptoms Medication Pneumonia Rules Generation Cough, Shortness of Breath, Fever Rules (Disease= Diabetes) => Chlorpropamide Amoxicillin, Augmentin

18 Rules (Disease=Heart Attack) => Aspirin Rules (Disease=Dengue and Symptom=Headache) => Acetaminophen Rules (Disease=Pneumonia and symptoms= Shortness of Breath) => Amoxicillin Rules (Disease=Pneumonia and symptoms= Chest Pain) => Augmentin Rules (Disease=Pneumonia and symptoms= Fever) => Tylenol It is apparent that ILA provides more range of results compared to Apriori Algorithm in terms of suggestions. 5. Conclusion This paper focuses on clinical decision support system for a virtual clinic which is a telemedicine proposal of healthcare service provision in remote areas of Pakistan. The idea is to connect the remote area patents with urban area hospitals/doctors using information and communication technologies. In addition to this, the system is equipped with clinical decision support system (CDSS) which helps the system to choose the appropriate doctor for a patient based on his previous prescriptions and a Fuzzy Rule Based System is augmented in the system to rank a doctor. CDSS also helps doctors in terms of suggestion of the medication for the patient based on his/her symptoms. This is accomplished by applying data mining techniques (Apriori Algorithm, Inductive Learning Algorithm) on system history (by the same doctor, by all the doctors) and online BNF database. Results show that inductive learning outperforms Apriori Algorithm in many ways. In future, further data mining techniques may be applied for better predictions. References [1] Stanberry, B (2001). Telemedicine: barriers and opportunities in the 21st century (Internal Medicine in the 21st Century). J Intern Med 2000, Vol. 247, pp [2] Sood, S., V Mbarika, S. Jugoo, R. Dookhy, C. R. Doarn, N. Prakash and R. C. Merrell What Is Telemedicine? A Collection of 104 Peer-Reviewed Perspectives and Theoretical Underpinnings. Telemedicine and e-health. October 2007, 13(5): doi: /tmj [3] Armstrong, I. J and W. S. Haston Medical decision support for remote general practitioners using telemedicine. [4] Carson, E. R Clinical decision support, systems methodology, and telemedicine: their role in the management of chronic disease. Information Technology in Biomedicine, IEEE Transactions, Vol. 2, Issue: 2, pp [5] Yuechun, C. and A. Ganz WISTA: a wireless telemedicine system for disaster patient care. Mobile Networks and Applications vol 12, issue 2-3, pp [6] Atta-ur-Rahman, S. Das, Big Data Analysis for Teacher Recommendation using Data Mining Techniques, International Journal Control Theory and Applications, vol. 10 (18), pp , [7] Atta-ur-Rahman, S. Das, Data Mining for Students Trends Analysis using Apriori Algorithm, International Journal Control Theory and Applications, vol. 10 (18), pp , [8] Atta-ur-Rahman, S.A. Alrashed, A. Abraham, User Behavior Classification and Prediction using FRBS and Linear Regression Journal of Information Assurance and Security, vol. 12 (3), 86-93, [9] Atta-ur-Rahman, Salam M.H., Jamil S. Virtual Clinic: A Telemedicine Proposal for Remote Areas of Pakistan, 3rd World Congress on Information and Communication Technologies (WICT 13), pp , December 15-18, Vietnam, [10] Memon, T. D., B. S. Chowdhry and M. S. Memon The Potential of Telemedicine System. An Approach towards a Mobile Doctor. National Conference on Emerging Technologies. [11] Malik, A.Z (2007). Telemedicine Country Report-Pakistan. E-Health Networking, Application and Services, 9th International Conference, pp [12] Karim, S Clinical Decision Support System Based Virtual Telemedicine. Intelligent Human- Machine Systems and Cybernetics (IHMSC), 2011.vol 1