Medical IT specialist Ben Kanagaratnam and his team have built the Bayes diagnostic model on six liver diseases and their symptoms. If necessary, more diseases and symptoms can be added by the user. The current model integrates the clinical data, collected from 314 patients, suffering from jaundice, who have been treated at the University College Hospital in Galway. All this data first has been introduced into an Excel worksheet. The Bayes formula then was applied to the data to calculate the probability of occurrence of the diseases taken into account the presence or absence of the symptoms.
To display the probabilities in graphic and numeric format, the spreadsheet was divided into four different sections. The trainee is expected to use the first section to enter the patient data for the best six symptoms by filling out "1" if the symptom is positive, and "0" when negative. The second section has a line chart, displaying the accumulative probability for each disease and the diagnostic effect of the presence or absence of the symptoms. The third section is reserved for the actual Bayes calculation. A formula was written in the first cell and copied to the rest of the cells.
The result in percentages shows the probability of occurrence of the disease. The figures are also displayed as a line diagram to visualise the probability value for the disease graphically. The fourth section includes the frequency distribution of all the symptoms for the six diseases. The disease names are entered in each row and the symptom values for the disease are introduced into each column. The students can fill in new symptom names, if they want, in order to recognise symptom patterns which have a high diagnostic impact on the diseases and to facilitate decision-making and clinical judgement.
The research team used the Microsoft Frontpage 2000 software programme to generate a home page on the Internet for the Bayes diagnostic model and exported the Excel worksheet with the calculated data and the results to the hyperlinked sub-directory. At Mednet 2000, Mr. Kanagaratnam explained that the Web-based model is easy to build, although its diagnostic accuracy does not surpass that of a professional clinician. Also, it has been set up as an education tool and as such, the model is not suitable to be implemented into the daily clinical practice.
One of the great advantages of the model being Internet-based is the smooth access to the diagnostic database at any moment. This provides tremendous opportunities for the students to communicate medical information. More information about the scientific work of Mr. Kanagaratnam is available in the VMW October 1999 article The value of accurate neural networks for future medical decision making.