In this work, we present a random number generator to generate virtual patient cases for a less common disease. The medical scientific literature provides diagnoses with the associated spectrum of symptoms and the respective probability of occurrence of each symptom [16]. Using brain abscess as an example, a Bernoulli experiment was performed for each symptom with the probability of success based on the literature data. A series of experiments for the symptom complex was started 10,000 times. Thus, 10,000 virtual patient cases with different symptoms were generated. Subsequently, the experiment was repeated 10,000 times under the same conditions. The result shows that the relative frequencies of the symptoms do not change significantly when the experiment is performed again. The generator can create virtual patient cases at each startup which are different in their symptoms. Although these are random, they are based on a Bernoulli distribution.
The main limitation of our generator so far is that specific symptoms are not sufficient to characterize a patient case. Additional information must be provided, this should include, for example, the following aspects:
- age
- gender
- origin
- socio-economic aspects
- further diagnoses
- further symptoms
- risk factors or predisposing conditions
In our literature research, we were able to find information on several of these aspects [20,25-27], and medical textbooks are also rich with specific information that could be implemented in an automated generation of patient cases [15,16]. Brain abscess, for instance, occurs more frequently in males (0.7/100.000) [28]. Our generator could be expanded to determine gender as well, including a new Bernoulli experiment with the probability of success being 0.7 for male gender. The information on gender can then be added to the constellation of symptoms. A further development of our generator can consider some of these other aspects in which patients differ. It would be of great benefit if a patient case with additional diagnostic criteria could be generated as a basic construct that would facilitate further elaboration. In the case of brain abscess, information on a predesposing condition like otitis media, sinusitits or heart disease would be desirable. These conditions, together with the range of their relative occurence, can also be found in the literature [28]. Moreover, a virtual patient should include laboratory data and media (like CT or MRT images), where necessary, as well as expert comments in the form of additional medical knowledge on a specific topic. For example, if there is a virtual patient with a suspected brain abscess, the expert comment “MRI is the first imaging choice for a patient with a suspected brain abscess. A lumbar puncture should be performed with caution only when there is clinical suspicion of meningitis or abscess rupture.” could be given according to the literature [20,29]. It is also possible that medical information is needed not only in binary (true/false) form, but in a quantitative form with numerical values. For example, for the symptom “fever”, in some medical contexts, the numerical value is needed (e.g. 38.5 degree Celsius). If this information is required, the authors of virtual patients would have to add the value manually. But for known distributions or ranges, methods of random generation of data can also be applied, such as those used in decision analytic procedures [30]. And even conditional probabilities could be simulated program controlled.
So far, the generator presented here does not provide any further information and manual editing of the generated patient case is necessary to add it. A more elaborated version of our generator could provide an extended construct that saves medical authors time which they can use in their clinical work, but it does not yet create a complete virtual patient.
Virtual patients and virtual cases are an integral part of medical teaching, especially in e-learning systems, but their development is expensive and complex [6,10]. Often virtual patients are based on real patient histories that are prepared for use in scenarios that are also virtual [10]. Little is known about the automated generation of virtual patient cases, and using statistical distributions of patient or disease characteristics seems to be a completely new field. Instead of using data from single real patients, we used statistical information on aggregated data as they are presented in textbooks or epidemiologic surveys. In this work we could take a first step in this direction and show, that it is possible to generate virtual training cases by performing Bernoulli experiments based on probabilities from the literature. So we could show that the research in this new field is possible and should be further expanded. This can be a useful benefit, as medical staff respectively medical teachers are very busy, and the automated creation of virtual patient cases saves them time. As a result, the medical teachers can spend more time with their real patients and more virtual training cases are available. Furthermore, a shortage especially of cases of rare diseases can be avoided. In a continuation of this work, better elaborated virtual training cases can be made available. This means that a constellation of symptoms and other data about a particular disease are presented, and the medical teachers can manually complete it into a virtual patient by adding further aspects such as expert comments, media and feedback. As a result, the education of medical students can be improved.