During statistical analysis of the data, abnormal and equivocal findings were both treated as abnormal results. An analyzer (ROC Analyzer, version 5.2, Richmond, Va) and a system (Statistical Analysis System, PC-SAS version 6.04, SAS Institute Inc, Cary, NC) were used for statistical calculations. Separate estimates of sensitivity, specificity, and likelihood ratios were calculated for each examiner and for each technique. Likelihood ratios were calculated as follows: (probability of positive method result in a subject with radiographic lung abnormalities/probability of positive method result in subject without radiographic lung abnormalities) = (sensitivity/1—specificity). The 95 percent confidence intervals for likelihood ratios were calculated as described by Simel et al. Pooled estimates of likelihood ratios were made using the Mantel-Haenszel test (SAS Proc Frequency). canadianfamilypharmacy
The magnitude of examiner effect was evaluated by including information from all three physical examination methods as the dependent variable in a receiver operator characeristic (ROC) curve analysis. An index was calculated on each case as the sum of the number of physical examination techniques that correctly identified radiographic confirmed lung abnormalities. Thus, for each examined lung, by examiner, a sum score was formed that could range from 0 to 3, depending on the number of techniques that found positive results. Areas under the ROC curves on this index were calculated for each examiner (by ROC Analyzer).
Interobserver agreement was assessed using the kappa coefficient. The kappa statistic ranges from +1 (perfect agreement) to — 1 (perfect disagreement). The kappa statistic only shows interobserver concordance and does not reflect the accuracy of the examination. Kappa coefficients were calculated (using ROC Analyzer).
Finally, combinations of examinations were looked at to calculate performance characteristics of each possible combination of techniques, given at least one positive finding for each combination. Of the patients offered the opportunity to enroll, 66 of 138 (47.8 percent) consented to participate and were assigned consecutive numbers. Of those who did not enroll, 34 (24.6 percent) were unavailable for examination, 22 (15.9 percent) were unwilling to participate, 8 (5.8 percent) had no chest radiograph performed, and 8 (5.8 percent) were unable to participate because of physical difficulties. Of the 66 patients who were originally enrolled in the study, 3 were excluded from the study because all physical exminations were performed greater than 48 h after chest radiographs were taken or no chest radiograph was found. Three of the patients were enrolled twice, each on separate hospital admissions, and were included as two separate examinations.