Background/aims: In the European Union Euricterus Project on (sub)Icterus proforma, the history and physical examination items were to be used for the physician's working diagnosis (PWD) and 'among others, for the development of the real life data electronic diagnostic tool, Trial. Trial delivers diagnosis probabilities based on Bayes' Theorem (B), completed by Trial Algorithm (TA). We wanted to compare the diagnostic accuracies (PWD and Trial probabilities as a percentage of the final diagnosis (FD) in a patient population) in 3 Dutch databases.
Methodology: The inclusion criteria for both Euricterus and Trial were age > or = 16 and bilirubin > or = 20 mmol/l. Euricterus data gathering took place at the bedside on a proforma with (among other questions) 79 questions on history and physical examination as well as the diagnosis levels for the PWD (1 alternative possible) and FD (17 disease categories, dc). Trial was developed on the data of 7,104 Euricterus patients and its data-entry Demo has the same questions. It calculates the probability of each diagnosis of the 17 dc as a percentage, as each significant finding is encountered (BO, Bayesian Overall). It can simultaneously calculate the resemblance of the patient's signs and symptoms to each disease concomitantly (BV, Bayesian Vertical), and to any subset of a disease. Any probability is further tested for compatibility using TA, a subset of BV, delivering TA-PWD, TA-BO and TA-BV. The Trial test patients came from 3 databases: a Euricterus Dutch Patients Random Sample EDRS (n = 184, internal database) and 2 independent databases: prospective P (n = 80) and retrospective R (n = 152), totalling 416 patients.
Results: The accuracies of PWD and Trial showed no differences between the databases, and the results are therefore pooled (n = 416). With testing on the highest probability found, the PWD accuracy was 78%, TA-PWD 81%, TA-BO 74% and TA-BV 72%. The true FD's were mentioned (at any probability) in the PWD in 86%, TA-PWD in 92%, TA-BO in 94% and TA-BV in 91% of the patients. Testing only patients whose FD was "certain" or whose data were without omissions did not improve accuracy. Testing on probability > 95% improved BO and BV accuracy, but not TA-BO or TA-BV.
Conclusions: The Physician's Working Diagnosis accuracy was approximately 80% and did not greatly improve after TA. The Trial TA-BO and TA-BV accuracies were only slightly less than the PWD. For well-trained physicians, Trial strengthens the physician's judgment, and for those less trained (or those to be trained), it delivers a (sub)icterus diagnostic disease probability at nearly consultant level.