A system for probabilistic diagnosis of jaundice has been used for studying the effects of taking into account the unreliability of diagnostic data caused by observer variation. Fourteen features from history and physical examination were studied. Bayes’ theorem was used for calculating the probabilities of a patient’s belonging to each of four diagnostic categories.The construction sample consisted of 61 patients. An equal number of patients were tested in the evaluation sample. Observer variation on the fourteen features had been assessed in two previous studies. The use of kappa-statistics for measuring observer variation allowed the construction of a probability transition matrix for each feature. Diagnostic probabilities could then be calculated with and without the inclusion of weights for observer variation. Tests of system performance revealed that discriminatory power remained unchanged. However, the predictions rendered by the variation-weighted system were diffident. It is concluded that taking observer variation into account may weaken the sharpness of probabilistic diagnosis but it may also help to explain the value of probabilistic diagnosis in future applications.
Modification of raindrop size distribution due to seeder–feeder interactions between stratiform precipitation and shallow convection observed by X‐band polarimetric radar and optical disdrometer
