AsthmaCritic's implementation is based on the generic critiquing model of HyperCritic.21
This generic critiquing model supports the integration of a critiquing system with an electronic patient record. HyperCritic's ability to process time-stamped data was limited. In addition, HyperCritic presented its output as text, often several pages long, with no opportunity for the physician to control the behavior of the system. Therefore, HyperCritic lacked structures that enabled integration of the system in the physician's working environment and was never tested in daily practice.
AsthmaCritic's implementation differs from HyperCritic's by (1) increasing the use of time-stamped data, (2) allowing the physician to control the output, and (3) supporting functions that allow a better integration with the electronic patient record (for example, running the system in background mode, attaching messages to a patient's records, adding the results of the analysis to the medical record, or monitoring what feedback still has to be shown to the physician).
The generic critiquing model distinguishes between critiquing knowledge (knowledge that initiates and guides the critiquing process) and medical knowledge (the medical base for critiquing). The analysis of patient data and medical knowledge is controlled by critiquing knowledge. Critiquing knowledge is described as a hierarchical set of critiquing tasks. Each individual critiquing task describes a specific procedure that defines for which clinical situations a critiquing statement should be generated.
Critiquing tasks are triggered by event descriptions
that identify the criteria that have to be met by the patient data for the task to be executed. Executing the critiquing task may result in a critiquing statement. , for example, shows a critiquing task that is executed whenever a drug is prescribed. This task searches for interactions between drugs. If such an interaction is found, the critiquing statement “possible interaction” is generated.21
Example of a critiquing task's specifications. If a clinical situation, as described by its event descriptions, matches a critiquing task's specifications, feedback is generated.
AsthmaCritic's critiquing knowledge is divided into four categories of critiquing tasks: diagnostic tasks, therapy-related tasks, referral-related tasks, and follow-up–related tasks. Each category is further subdivided into more specific tasks. For example, a therapy-related task contains nine different subtasks (such as checking for contraindications, checking for drug interactions, verifying dosing schedules). In total, AsthmaCritic contains 131 specific tasks.
Critiquing tasks, however, are solely procedural specifications. Based on the existing medical knowledge, the total number of different clinical situations that can be distinguished is much higher. For example, screening for contraindications is a single task, but the number of clinical situations that will be detected depends on the number of drugs and contraindications known in present-day medicine. As such, critiquing tasks only specify the process of a critique, not the content. For the critiquing statement to be generated, medical knowledge needs to be available (for example, dosing schedules, side effects, interactions).
AsthmaCritic's medical knowledge base has been built from national guidelines,22,23,24
pharmaceutical reference books,25
guides on interactions and side effects,26
and the existing consensus among a group of specialists in asthma and COPD. Building the knowledge base was a three-year iterative process under the guidance of a medical content board consisting of four local specialists (two general practitioners and two pulmonologists) and seven national specialists. Members of the medical-content board reviewed each new version of the knowledge base.†‡