Decision support for the differential diagnosis of jaundice

Making a diagnosis is always a challenge for the clinician who faces a jaundiced patient. A diagnosis provides a basis for prognosis, and guides therapeutic action. Modeling of the diagnostic process provides an understanding of the contributing elements. In addition, diagnostic models can be applied to real world situations. This can be useful when the model supports or complements the clinician on particular tasks. In that way, the diagnostic model becomes a diagnostic aid. Computers can often be successfully used to implement diagnostic models, hence the term 'computer assisted diagnosis'. The performance of diagnostic aids is always an important issue. Most diagnostic aids are developed at one place, whereas they are used at an other place. In the development of diagnostic models, one often models features that are characteristic to the local patient population. These features may be different for patients elsewhere. There is a chance that there will be a decline in performance of the diagnostic aids due to these differences, and that adjustment procedures are required. These are the crucial questions addressed in this thesis. The symptom jaundice is a good starting point to model diagnostic processes. There is a reasonable consensus regarding the diagnostic categories. Diseases underlying jaundice can have serious consequences for the patient. Jaundice also is a relatively frequent symptom. These properties, together with practical, historical and organizational motives explain our choice of jaundice. Two retrospective data collections on jaundiced patients admitted to the department of Internal Medicine of Dijkzigt hospital are used throughout this thesis. A first series included 100 patients, whereas a second contained 214 patients. With these data collections at hand, we could answer the questions formulated. Especially the second data collection provided detailed information on the diagnostic parameters of the patients. There we also recorded the referral patterns. These data confirm that Dijkzigt hospital operates as a hospital with many secondary and tertiary referrals. The implementation of the data collections will be outlined in chapter 2, whereas chapter 3 present some background data on the patient populations.