Predicting enteric fever without bacteriological culture results

We used Bayes' theorem to calculate the probability of enteric fever in 260 patients presenting with undiagnosed fever, without recourse to blood or stool culture results. These individuals were divided into 110 patients with enteric fever (63 culture positive, 47 culture negative) and 150 patients with other causes of fever. Comparison of the frequencies of occurrence of 19 clinical and laboratory events, said to be helpful in the diagnosis of enteric fever, in the two groups revealed that only 8 events were significantly more frequent in enteric fever. These were: a positive Widal test at a screening dilution of 1:40; a peak temperature greater than = 39 degrees C; previous treatment for the fever; a white blood cell count less than 9 X 10(6)/litre; a polymorphonuclear leucocyte count less than 3.5 X 10(6)/litre; splenomegaly; fever duration greater than 7 d; and hepatomegaly. When the probability of enteric fever was determined prospectively in 110 patients, using only 6 of these discriminating events, the probability of patients with a positive prediction having enteric fever (diagnostic specificity) was 0.80 (95% confidence interval: 0.68 to 0.91) and the probability of those with a negative prediction not having enteric fever (diagnostic sensitivity) was 0.92 (0.85 to 0.99). Using all 19 events did not alter the diagnostic specificity or diagnostic sensitivity. This study shows that a small number of clinical and laboratory features can objectively discriminate enteric fever from other causes of fever in the majority of patients. Calculating the probability of enteric fever can aid in diagnosis, when culturing for salmonella is either unavailable or is negative.