Nowadays, technology is growing rapidly. With such tremendously growth of technology, many field of industry is taking the chance in adopting these technologies to transform their business flow to fit with the environment. Medical is one of the industries that changing their services to provide better care and better treatment to patients. Many clinical center, hospitals or medical organization is investing on Clinical Decision Support System to improve the quality of decision making from the progress of diagnosis.
What is Clinical Decision Support System?
Clinical Decision Support Systems are “active knowledge systems which use two or more items of patient data to generate case-specific advice” from Wyatt J, Spiegelhalter D, 1991 (OpenClinical 2001-2009)
It designed to integrate with a medical knowledge database as well as patient data to generate case specific advises to users. In another words, it is designed to healthcare professional to make medical decision.
Instead of taking the place of diagnosis as a job of computer program, it rather intended to support the clinical experts because computer is not able to perform as a human being and it may cause error which may harm and risking others people survivability.
In some area, computers can help the clinician in retrieving details needed in the progress of diagnosis such as patient’s medical history, all kind of examination and laboratory test. In addition, the reaction of drug and allergies toward the patient will be taken into account to help a busy clinician to handle over hundred patients in a day. (Clinical Decision Support System, Citizendium, 2006)
What is the purpose of Clinical Decision Support System?
CDSS generally is used to assist clinician by using the point of medical to provide some expert opinion or advices. A clinician may interact with CDSS in doing determination of diagnosis, analysis and etc by according to provided patient data.
Previous theories of CDSS were to use the CDSS to literally make decisions for the clinician. (Clinical Decision Support System, iScanMyFood, 2010). By now, clinician is able to input information to the system and wait for CDSS to output the right choice to advice them the correct action.
By gone through the computer analysis, clinician is not only making decision through own knowledge which may not be most suitable result from a diagnosis but also getting advices from computer to improve the quality of decision making. In another words, it served as a peripheral brain.
Functions of Clinical Decision Support System
There are 4 basic functions contain in Clinical Decision Support System which are Administrative, Managing clinical complexity and details, Cost control, Decision support by based on Perreault & Metzger.
Administrative means system must be administrable which means that it must be able to support clinical coding and documentation, procedures and referrals of the medical center. In order to achieve that, CDSS is always created through multiple platforms and it understands very well on every medical’s standard procedure.
Other than that, it must be able to manage clinical complexity and details. It keeps patients on research and chemotherapy protocols as clinical experts always did. It tracks patient orders, referrals follow-up the status of patient and preventive care after prescription.
Cost controllable by avoiding any duplication of process, document or any unnecessary lab test and to monitor medication orders to confirm any incorrect places which might be a direct harm to particular medical center’s financial
Decision Support is mean to support clinical diagnosis and treatment plan processes and promoting use of best practices, condition-specific guidelines, and population-based management. (OpenClinical 2001-2009)
Characteristics and Types of Clinical Decision Support System
Characteristics of CDSS
There are 4 basic component usually required by CDSS which are Inference Engine, Knowledge Base, Explanation Module and Working Memory.
Inference Engine is the main part of CDSS. It used knowledge from database integrated with the system as well as the knowledge about the patient to generate an output or a conclusion based on certain condition. Inference engine control the actions of the system and guide system with the best actions. For an example, it will start to detect the condition to trigger the alert or conclusion to be displayed in a diagnostic progress.
Knowledge Base acquired the knowledge Inference Engine used to present to the users. In Knowledge base, it contains every risk factor to carry out in new lesions and risk scores. It will be built with the involvement of clinical domain experts with also every activity of create, edit and maintenance. In another way, some knowledge base is created through automated process. Automated process knowledge is acquired from external sources such as books, magazine, journal articles and database by a computer application. The process of creating a knowledge base is complex and complicated. In order to make it easier, there are tools specially created to facilitate the acquisition and elicitation of knowledge base. There is an example tool called Protege, a knowledge- based development environment.
Working memory is a collection of patient data or form of a message which is stored inside database. These data may include patient’s age, name, data of birth, gender and etc or allergies, history medical information or problems and other information.
Explanation Module responsible in composing justification for the conclusions drawn by the Inference Engine by applied Knowledge base and patient data. This component is not presented in all CDSSs.
In another way, CDSS can work on synchronous mode and asynchronous mode. In synchronous mode, users can communicate directly with application to wait for the output from system. Users will have to wait for the output in order to continue their works. For example, CDSS checks for drugs interaction or any possible medicine that patient allergies to then clinician will only able to continue to diagnose patient by based on the result generated by CDSS. When there is in asynchronous mode, CDSS is performing independently while does not required user to wait for. For an example generate a checkup reminder for patients.
CDSS can be categorized as open-loop or closed-loop systems. Open-loop CDSS will generate a conclusion but it takes no action directly by its own. Usually users will take the actions on the final decision. For an example, CDSS generates alert or reminder to users to take the actions. A Closed-loop CDSS is the opposite of open-loop CDSS. It will take actions by its own without any intervention from users. For an example, system will automatic save up all details of diagnosis process.
CDSS can be also an event monitor, a consultation system or a clinical guideline. Even monitor is a software application that converts every available data into electronic format and uses its integrated knowledge base to send reminder to clinicians appropriately. Consultation system allows user enters the details of a case and in another way, the system will provide user a list of problems that may explain the case and suggestion the best action to be taken.
Clinical Guideline basically developed by a group of clinical experts and disseminated by the government or by professional organization and it apply in most of the CDSS. This clinical guideline has been presented with every statement of best practices regarding to a particular health condition. Other than providing recommendation from various practices, it can be taken as examples in medical education.
Type of Clinical Decision Support System
Knowledge-based Clinical Decision Support System (Expert System)
Knowledge-based expert systems are created by having experts use the biomedical literature to identify relationships between independent variables (such as signs and symptoms) and dependent variables (such as likely underlying diseases).
It contains related arranged such as local hospital information, patient data and other compiled data and apply it with IF-ELSE-THEN predefined rules to guide through the whole progress of decision making. However, rules may also be acquired from various types of decision trees.
These rules-based CDSS is the most usually found among all the clinical application. It will alert user when there is a possible drug doses or allergies which may harm or risk patient life by based on patient details such as age, sex, weight, height and etc.
Example: if the system rules used to determine drug interaction, the formula will started to run and to detect every possible risky drug interaction, the rules might be IF drug A is taken AND drug B is taken THEN alert user. By going through these predefined rules, provided information must be always up-to-dated to prevent any wrong output which might lead to misdiagnosis.
To construct a rule-based system for medical decision support, an expert with domain knowledge always must be recruited to create and handle the knowledge base and train the system. To train an expert system is very time-consuming and it the result that produced is only usable in a narrow scope project. Therefore, a rule-based CDSS is not commonly used to deliver the critical message to clinician. (Clinical Decision Support System, Citizendium, 2006)
Non Knowledge-Based Clinical Decision Support System
Non Knowledge – Based CDSS does not apply any data from knowledge base but they used another kind of artificial intelligent called Machine Learning. From the term of Machine Learning, it means a machine will learn from the past experience and previous lesson that given by experts. This kind of idea has implemented in this type of CDSS. Computer will learn everything in previous medical progress and find pattern in clinical data.
Non Knowledge – based CDSS is trained from the relationship between symptoms and signs (also called independent variables) and diseases (also called dependent variables). Machine Learning is using case-based to proceed every lesson because the system is being trained from previous cases.
There are 2 type of non knowledge – based systems are artificial neural networks and genetic algorithms. It contains some mathematical models that can observe and emulate the properties of an item and some kind of adaptively learns the simulated properties of the item. (Clinical Decision Support System, Citizendium, 2006). Artificial neural networks type of CDSS can analyze the attributes or patterns from patient data to derive the associations between the symptoms and a diagnosis. (Wikipedia, 2010). It can perform supervised or unsupervised machine learning depending on the way of providing the available information.
Genetic Algorithm is based on a several processes of searching and simplifying and use the directed selection achieve optimal CDSS result. The algorithm will first determine properties of sets of solutions to a problem. Every solution that generated will be recombined, mutated and repeat the process again. The rotation of finding solution will not stop until a proper solution is found. The knowledge used in finding solution is derived from patient data. It usually focus on those disease that caused by narrow list of symptoms. (Wikipedia, 2010)
Architecture of Clinical Decision Support System
3.1. Basic Concept of Decision Support System Architecture
Since Clinical decision support system is a kind of decision support system that is design to assist clinician in decision making tasks. The architecture design of decision support system always consists of two major sub-systems which is human decision maker and computer systems. Construct a decision support system with only computer hardware and software program is not a correct concept because there might be some unstructured or semi structured decision (those decisions cannot be decide through a collection of mathematical model or formula) is not able to be programmed by system because it’s precisely nature thinking from a human and it is elusive and complex. There is no such independent component in a decision support system. It always needs a human decision maker as another component of decision support system to integrate with computer systems. The function of human decision maker is not to build a database for decision support system. Instead of build a database, it functions as a “decision maker” that provides judgment, share their experience and exercises intuition throughout the entire process of decision making.
The very first step of decision making is begin with the creation of a decision support model (decision support model is the formula or the way that helps user to filter or decide the specific result) by using some integrated DSS program such as Microsoft Excel. System will interact with database through Database Management Systems (DBMS) and deal the data from database with the decision support model through Model-Based Management System (MBMS). DBMS is an application that used to create, manage as well as control the access to the database. MBMS is an application that embedded within a DSS program that allow user to create, edit and delete the decision support model. By going through DBMS and MBMS, model is able to associate with the data from database to make a specific decision.
Figure 1.0 Decision Support System diagram
The diagram above shows DBMS and MBMS is integrated with the DSS to communicate with the models and database to provide result to users.
3.2. Four-Phase Model of Clinical Decision Support Architecture
Four-Phase Model of clinical decision support architecture is referring to 4 type of architecture that has been used in clinical decision support system development. These architectures also representing the evolutionary of clinical decision support system. This 4 type of architecture is standalone decision support system (1959), integrated system (1967), standards-based system (1989), service models (2005). The phases is happen sequentially, every phase is learned and influenced from previous phases.
Standalone Decision Support System
The first phase is Standalone decision support system which happened in year 1959. They were systems that operate separately from clinical system. The clinician got to purposely seek the system out and enter information of his medical cases and then wait for the system to interpret the result. This kind of system is easy to develop because user that comes with medical knowledge and computer skills can make one of it. It is easy to share as well because the system is easy to develop, it can be categorized as a simple system, user can just make a copy of the program and then mail to another who wishes to use the system. There are limitations such as they required user to enter all the information needed by the system to make it inference. Another disadvantage is user got to seek out how the system works and flow. User that is lack of medical knowledge might have problem in system usage and might causes a lot of medical error. Thus, they cannot be proactive. It also very time consuming, it may takes half to an hour to enter a case because the model’s feature is very narrow and it required a lot of information to generate an output.
Due to the significant problems from standalone CDSS, developers begun to involve the architecture into another which is integrated system. The invented of Integrated system have solved a lot of problems. First of them is termination of multiple user input. The information is stored electronically after the first input by the user. Another significant solution is system can be proactive. They can alert user when it detect dangerous between drugs interaction or the dosing error automatically. The major disadvantage of integrated system is difficult to share. This system is very complex because it directly built with large clinical system. Therefore, it can’t directly share to others who are not using the same clinical system. Unlike standalone system which built only based on self knowledge and computer skills. It can be send to anyone who wanted to use it. Another major problem is knowledge management problem. When there is an update for knowledge or clinical guideline, it maybe needs to find the source code to know where is guideline used.
In order to make content sharable, several research and effort had been undertaken to standardize clinical decision support content. The standardization of content has overcome many disadvantage of integrated system. It shares the clinical decision support content by separate the code that describing the content from source code. However, it still has some limitations. First, there is way too much standard format to choose. There are over hundred of standard to represent a simple notification. Standardized encoded may constrain a user’s standard. The “standard” that user intended to write has the difficulty to compatible with the “standardized standard”.
Service Models, the most recent CDSS architecture. It recombined clinical information system and clinical decision support system components by using a standard application programming interface (API). This models standardizing both clinical decision support system and clinical system into one interface. Both systems will only look at only one clinical system and one CDSS at a time although the knowledge about patient and medicine are across many places.
Clinical Decision Support’s Algorithm
4.1 Artificial Neural Network
Artificial Neural Network is a method that used by non knowledge-based CDSS. It required training from experts in a form of artificial intelligence. It will base on the past experiences or recognized examples to create a set of solution to a medical problem. They possess the “Human-Brain-Like” behavior instead of “Computer-Like”. Due to the capability of knowing the “behavior” of problem through its experiences, they are commonly used in recognition problems. From the result, this methodology is very well in determining narrow and well-defined clinical problem.
Three general type of algorithm used by machine learning which is unsupervised, reinforcement and supervised.
Unsupervised learning means the computer identify some natural grouping within a database by based on how “similar” the items are and what makes a “Good” group without being provided examples of feature values of items. Therefore, the way of machine learning also called “clustering”. Unfortunately, unsupervised learning is not being used in many studies of various type of diagnosis.
In reinforcement learning, it is not provided any samples of feature values of items. Instead of giving the samples, it is given a specific main point or feedbacks which are able to determine whether the system is on the right track.
In supervised learning, the computer is given the samples of feature value of items. The reason of doing supervised learning is to develop a “classifier” that can predicts all the possibility from given predetermined classes or samples based on a set of attributes and features to describe the items.
4.2 Bayesian Network
Bayesian Network shows a set of variables and dependencies of conditional among the variables via Directed Acyclic Graph (DAG). Each node in the graph represents a variable and particular node will link to its neighbor to show the dependencies among the corresponding variables. This algorithm provides a simple understanding and definition between any two nodes. It helps predict and compute every possibility event might occur in a specific condition. In the stand of medical view, it can compute every possibility diseases by based on the symptoms given. For example, fever, cough, sore throat and chilling might lead to symptoms of Dengue disease.
There are two important component consists in this algorithm which are structure and a set of parameters. Structure of the Bayesian Network is constructed from DAG. Every node in DAG may be given value by the parent node. Parameters are describing the relationship and the probabilities of a node to its parent. These components can support Bayesian Network computation by using the chain rule. Therefore, parameter and structure learning must be carrying out to fully represent probability distribution. Parameter learning is to specify each node in DAG is approximately distributed based on varies conditional. Structure learning is to identify the way of distribution throughout the whole network by based on the local data.
When learning Bayesian Network, the amount of training data is very important and it directly affected the correctness of the network. Therefore, training data must be provided enough through employment of experts to provide various form of knowledge to improve the accuracy of the models. The experts might provide some knowledge that specifying a condition among the variables in Bayesian Network.
Bayesian Network Example.png
Figure 2.0 Example of Bayesian Network
The example shows that fever and chilling maybe the symptoms of Dengue Disease. In another way, chilling maybe the side effect of fever.
4.3 Logical Condition