Application of probability analysis in the diagnosis of coronary artery disease

Comparison of logistic regression and Bayesian-based algorithms to estimate posttest probability in patients with suspected coronary artery disease undergoing exercise ECG

Two multivariate methods, a logistic regression-derived algorithm and a Bayesian independence-assuming method (CADENZA), were compared concerning their abilities to estimate posttest probability of coronary disease in patients with suspected coronary disease. All patients underwent exercise testing within 3 months prior to coronary angiography. Coronary disease was defined as the presence of one or more vessels with greater than or equal to 50% luminal diameter narrowing. A group of 300 patients (disease prevalence = 37%) was used to derive the algorithm. Another group of 950 patients was used to validate the algorithm and compare it to CADENZA. Seven variables (age, sex, symptoms, diabetes, mm ST depression, ST slope, and peak heart rate) were used to generate posttest probabilities for each method. The receiver operating characteristic curve area for the logistic regression method (0.81 +/- 0.01) was significantly higher than CADENZA (0.75 +/- 0.01; p less than 0.05). There was, however, no difference in the calibration of the two methods. When given equivalent variable information, the logistic regression algorithm had better discrimination than CADENZA for estimating the probability of coronary disease following exercise electrocardiography.

The estimation of post-test probability of coronary disease following exercise testing using the sequential application of two Bayesian methods

Recent studies have revealed that Bayesian methods to estimate post-test probability following exercise testing differ in their sensitivity and specificity across the range of post-test probability. To take advantage of the relative strengths of each method, we combined two of these methods into a single method (DUAL BAYES) and compared it with the two original methods in 436 patients who underwent stress testing followed within 2 months by coronary arteriography. All patients had post-test probabilities determined using CADENZA (better sensitivity). Those CADENZA-derived probabilities greater than or equal to 50% were substituted with post-test probabilities determined by Diamond and Forrester's original TABULAR method (better specificity). Mean post-test probabilities were as follows: TABULAR 34, CADENZA 48, DUAL BAYES 37 (actual incidence 38%). Comparison of sensitivity and specificity at every fifth percentile of post-test probability revealed that the sensitivity of DUAL BAYES was better than that of TABULAR and equal to that of CADENZA at thresholds less than or equal to 10 and that the specificity was better than that of CADENZA and equal to that of TABULAR at thresholds greater than or equal to 60. Therefore using both methods as indicated above was better than using either method alone.

Comparison of three Bayesian methods to estimate posttest probability in patients undergoing exercise stress testing

To determine whether recent refinements in Bayesian methods have led to improved diagnostic ability, 3 methods using Bayes' theorem and the independence assumption for estimating posttest probability after exercise stress testing were compared. Each method differed in the number of variables considered in the posttest probability estimate (method A = 5, method B = 6 and method C = 15). Method C is better known as CADENZA. There were 436 patients (250 men and 186 women) who underwent stress testing (135 had concurrent thallium scintigraphy) followed within 2 months by coronary arteriography. Coronary artery disease ([CAD], at least 1 vessel with greater than or equal to 50% diameter narrowing) was seen in 169 (38%). Mean pretest probabilities using each method were not different. However, the mean posttest probabilities for CADENZA were significantly greater than those for method A or B (p less than 0.0001). Each decile of posttest probability was compared to the actual prevalence of CAD in that decile. At posttest probabilities less than or equal to 20%, there was underestimation of CAD. However, at posttest probabilities greater than or equal to 60%, there was overestimation of CAD by all methods, especially CADENZA. Comparison of sensitivity and specificity at every fifth percentile of posttest probability revealed that CADENZA was significantly more sensitive and less specific than methods A and B. Therefore, at lower probability thresholds, CADENZA was a better screening method. However, methods A or B still had merit as a means to confirm higher probabilities generated by CADENZA (especially greater than or equal to 60%).(ABSTRACT TRUNCATED AT 250 WORDS)