Clinical Decision Support System Computer Science Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

As cited in AMIAs CDS Roadmap, Clinical decision support provides clinicians, staff, patients or other individuals with knowledge and person-specific information, intelligently filtered or presented at appropriate times, to enhance health and health care. It encompasses a variety of tools and interventions such as computerized alerts and reminders, clinical guidelines, order sets, patient data reports and dashboards, documentation templates, diagnostic support, and clinical workflow tools.[41]" The development of clinical decision support system is one of the key points in the prevention of medical errors. It has been reported that if used properly, CDSS in reality has the potential to enhance the health care system [42]. Today, there are a variety of CDSS available in the market using different kind of methods and algorithms to support clinical decisions. Generally, the user will enter inputs like signs and symptoms, laboratory results, patient's information and many more into the system, then CDSS will intelligently running through all the inputs and eventually giving out appropriate suggestions. Typically, there are two types of CDSS which are knowledge-based and nonknowledge-based CDSS.

Generally in a knowledge-based CDSS, the system consists of three main components which are medical knowledge base, inference engine and a mechanism to communicate with the user [43]. The medical knowledge base comprises a collection of expert medical knowledge usually in the form of IF-THEN rules. Inference engine is where the reasoning mechanism occurs. The engine contains computer algorithms used to process the relation between the knowledge base and patient's data.

Figure 2.5: A general model of CDSS

The communication mechanism is a method to enable the user to key in the input and also display the result to the user. The input can be obtained from many different kinds of channels. In a stand-alone system, normally the user needs to enter the patient data directly to the system. In a larger system where CDSS is incorporated into electronic medical records (EMR) system, the patient data is accessed from the database where the health care providers have previously saved the data in electronic form [44]. Figure 2.5 shows a general model of CDSS.

The difference between a knowledge-based CDSS and nonknowledge-based CDSS is that the nonknowledge-based CDSS applies the principle of machine learning, a form of artificial intelligence, where it trains the computers to learn from past experiences and/or find patterns in clinical data without any medical knowledgebase stored in its system. The nonknowledge-based CDSS has two kinds of approach which are artificial neural network and genetic algorithm.

2.3 Clinical Diagnosis Reasoning Method

Human reasoning is a very complicated process and not only based on an exact logic. A doctor makes a diagnosis decision based on his knowledge on multiple principles and his long-term experience on the field. In the literature, many researchers have introduced different kinds of reasoning methods. Some of the reasoning methods are:

Logical conditions [45] - Logical reasoning is a simple and straight forward method. This method gives off an action based on a given variable and a bound. The reasoning process will check if the variable is within or outside of the boundaries and make a decision based on the result. For example: "is the patient's heart rate below 50 BPM?" Logical conditions are typically used to provide reminders and alerts to the health care providers.

Ruled-based systems [45] - The knowledgebase of this system is constructed based on the IF <condition> - THEN <action> rule statements. For example: "if the patient has a history of stillbirth, then she is at high risk of preeclampsia." Rule base is a compilation of these rules. The data can be obtained at the beginning of the process or it can be obtained by asking questions to the user, then the data will run though a chain of rules to eventually come out with a conclusion. The process of running through the rule base is called pattern matching, the pattern matching can occur in a forward-chaining, backward-chaining and mixed-chaining. The advantages of this system are it is very close to natural language and matches the natural reasoning process of a human being. The disadvantages are it is very hard to translate the medical knowledge into rule one by one exactly and many rules may be needed for the system to be effective. One of the most popular and also the first expert system using this approach is MYCIN, a rule-based system used to identify the infection caused by different kinds of bacteria.

Bayesian network - Bayesian network is a typical knowledge-based decision making system. Basically, Bayesian network is a directed acyclic graph representing the probabilistic relationships between a set of variables - disease and symptoms based on conditional probability according to the Bayes theorem [46]. The Bayesian probability theory helps to calculate the probability of a disease given the symptoms and also able to update the belief as new evidence come in. It is an advantage that the probability of a disease can be counted nevertheless the difficulty of getting the accurate probability of a disease has become its disadvantage [45]. DXplain is a CDSS that utilized this method to list out ranked diagnosis related to the symptoms.

Artificial neural network (ANN) [45] - ANN is a non-knowledge-based adaptive CDSS that uses the concept of machine learning where it can learn from the past experience and recognized patterns in the clinical information. ANN is made up of nodes called nodes called neuron and weighted connections that transmit signals between the neurons in a unidirectional manner. The advantage for an ANN system is the ability to build the knowledge without programming the system and providing inputs from the users. The drawbacks of this system are the time consumed for the training process and the reasoning behind the system could not be explained that makes the user to doubt the reliability of the system. Various applications have been using ANN such as the diagnosis of appendicitis, back pain, myocardial infarction, dementia, skin disorders and the prediction of pulmonary embolisms.

Genetic Algorithms - Another nonknowledge-based method in CDSS which is similar like ANN, they both use the patient data to build their own knowledge. The algorithms go though iterative processes to rearrange itself until it find optimal solutions based on the patient data provided. The fitness function is used to determine the good solutions and eliminate the poor solutions. Although this method is similar to ANN, it does not have many examples of CDSS compared to those using ANN.

Figure 2.6: Methodologies and technologies of CDSS

Over the years, various methodologies and technologies have been introduced in the field of CDSS. Some of the important methodologies and technologies of CDSS are information retrieval, evaluation of logical conditions, probabilistic and data driven classification or prediction, heuristics modeling and expert systems, calculations, algorithms, and multistep processes, and associative grouping of elements.

Information Retrieval

The ability to find medical related basic information is the central of a CDSS. A retrieval task can be as simple as searching for related medical information for instance the normal range based on the laboratory test and then proceed to a higher level where the system is able to provide the suitable suggestion regarding the result of the laboratory test. There are two types of techniques in retrieving the information which are taxonomy-based or ontology-based search and text-based search.

Evaluation of Logical Conditions

Logical conditions are the most popular techniques used in CDSS and plenty of logical conditions models have been explored. One of the main subjects that had been primarily worked at is refining and reducing the numbers of diagnostic possibilities which can be achieved with decision tables. Venn diagrams are also another popular way to illustrate the clinical logic. Logical expressions with Boolean combinations of terms are another common way to represent the logical conditions. Alert and reminders in this case are used to evaluate the triggered conditions when the inputs are given.

Probabilistic and Data-driven Classification or Prediction

Since most of the clinical diagnosis is not done with only a single precise aspect, the CDSS needs to recognize various types of medical data. In this section, the primary theory on probability of medical diagnosis is based on Bayes theorem. Developing decision tree is the core of standard decision analysis. Decision tree helps to outline the sequences of decision and possible effect of each node which the decision maker can focus on the important variable afterwards. Data mining is a database technology that slowly evolves in order to search for new valuable discovery or hypotheses. The development of ANN in the recent years is quite fruitful and ANN has become stronger in the field once again.

Heuristics Modeling and Expert Systems

Another methodology to emulate human expertise and reasoning process is rule-based systems. The fundamental of a rule-based system is the rule base composed from a set of rules in the form of production rule IF-THEN. Another heuristic modeling approach is the use of frame-based representation.

Calculations, algorithms, and multistep processes

Calculations, algorithms, and multistep process are usually unavoidable in a CDSS. They are used in computational processes executions, flowchart based decision logic, interactive user interface control, biomedical imaging, and image processing, and signaling.

Associative Groupings of Elements

The last section explains the approach in managing the data for presentation. The ket developments are report generators, business intelligence tools and document construction tools, document architectures, document and report templates, mark-up language, ontology tools and ontology languages. These elements are used for structured and relational data, structured reports, order sets, other specialized data views, presentations, business views, and summaries.