Probability Theory in the Diagnosis of Cushing's Syndrome

In the differential diagnosis of patients in whom Cushing's syndrome is suspected, the physician uses clinical signs and simple laboratory data in addition to information gained from past experiences to make a decision concerning the probability of the diagnosis and the need for further investigation. He usually does not make an explicit formulation of either the information or the reasoning process leading to his decision. This report is an examination of an explicit method for estimating the probability of Cushing's syndrome in patients in whom this diagnosis is suspected, using clinical signs and simple laboratory techniques, not including steroid analyses. The clinical and nonsteroid laboratory data of 211 patients investigated for Cushing's syndrome were examined to determine the incidence of the signs of the syndrome in the patients with and without the disorder. These data were used to calculate the probability of Cushing's syndrome in 111 additional patients examined because Cushing's syndrome was suspected. The diagnosis of Cushing's syndrome was established by steroid investigations, operative or autopsy findings, or the response to therapy in each of the 111 patients. The probability calculations using only clinical and nonsteroid laboratory data led to a confident diagnosis (p≥0.99) in 9 of the 38 patients in whom the diagnosis of Cushing's syndrome was established. Of the 73 patients who did not have Cushing's syndrome the calculated probability of Cushing's syndrome was 0.01 or less in 45. In other words, in one half the patients suspected of Cushing's syndrome, the diagnosis could be confirmed or excluded with a high degree of confidence and with an accuracy far greater than that provided by simple steroid screening tests such as single plasma or 24-hr urine 17-hydroxycorticosteroid determinations. In the remaining patients, less certain and less accurate predictions of the diagnosis could be made.