Computer-Assisted Decision Making

Approaches

A decision support software system has several conceptual components:

• 1.
  An inferencing approach typically embodied in an algorithm that enables the system to interpret patient data based on the knowledge available to the system. Examples of such approaches include Bayesian inferencing and production rule evaluation.
• 2.
  A knowledge base that comprises the biomedical knowledge available to the system. The knowledge is encoded in a form that corresponds to the inferencing approach being used. For example, a knowledge base for trauma diagnosis might consist of a Bayesian network relating patient symptoms and findings to internal organ injuries.
• 3.
  Optional interfaces to other computer systems to obtain data about a patient.
• 4.
  A user interface to interact with the user, such as to obtain data, and to present the results from the inferencing.

Based on the mode in which the decision support is invoked, the system may be characterized as one providing solicited advice or one providing unsolicited advice. In the former case, a clinician may seek recommendations from a decision support system to assist with making a differential diagnosis in a patient with an unusual presentation. Such systems usually contain a large knowledge base that spans a domain of clinical interest such as internal medicine or infectious diseases. Unsolicited advice is rendered by systems (a) in response to clinical events that are being monitored, such as the reporting of a critically low value for a serum potassium test, or (b) as a critique of a proposed physician intervention such as prescribing a medication to which the patient is hypersensitive. Decision support systems that offer unsolicited advice, to be able to function, must be integrated with sources of patient data such as an electronic medical record (EMR) system or a computer-based provider order entry (CPOE) system. Systems that offer solicited advice may be integrated with sources of patient data or may be freestanding.

An important aspect of decision support systems for clinical use is how it integrates into the clinical workflow. In other words, the successful use of these systems depends on when and where the system's advice is presented to the clinicians. Thus, various kinds of tools have been created to present advice at particular points in the clinical workflow. For example, reminder systems are used often to advise clinicians in the ambulatory setting about preventive care actions that are applicable to a patient. Electrocardiography (ECG) machines incorporate features to analyze the ECG and print or display the resulting interpretation of the findings with the ECG trace. Rule-based systems critique physician orders and prescriptions in the CPOE application to prevent orders that might have the potential to harm the patient or those that might be ineffective. Such systems also might suggest additional orders called corollary orders: For example, an order for a nephrotoxic medication might lead to a corollary order for performing kidney function tests. Abnormal laboratory test results are highlighted on the screen to draw the attention of the clinician to those values. Furthermore, links to didactic informational resources, also known as infoButtons, can be shown next to the results. These information resources can be used by the clinicians to help interpret the test result and decide on an appropriate action. Treatment planning systems for surgery or radiation therapy are used in a laboratory setting initially to plan the treatment. The outputs of these systems are presented to the clinician during the treatment in the operating room.

...