Is it possible to make an accurate diagnosis based only on a medical history? A pilot study on women's knee joints

The purpose of this pilot study was to determine if an accurate diagnosis could be made concerning the knee joint using only the patients' medical history information. Only women were chosen for this study because of existing unpublished data on a cohort of 100 women with normal knees to act as a control (group I). From the 2,266 knee surgical procedures in the database of one surgeon, two other groups were selected. Group II was those women with only a torn medial meniscus. Group III were those women with only a torn anterior cruciate ligament (ACL). The medical history data of one half of the database were statistically analyzed to determine the questions that were the best predictors of each group. The medical history questions discovered to be best predictors were different from what might be expected from an individual surgeon's experience, expert opinion, or a medical consensus opinion panel, but the predictors did have a foundation in fact and are substantiated by statistical analyses. Using these predictors, a validation was performed on the other half of the database. When the top 142 predicting questions were used, the diagnostic accuracy was 98%; 98 of 100 of the “normal” group, 57 of 59 cases classified as having a torn meniscus, whereas 128 of 129 cases classified as having a torn ACL were correctly identified. When the only the 30 strongest predictors were used, the diagnostic accuracy was 85%: 100 of 100 cases were correctly classified as normal, 45 of 59 cases were correctly classified as having a torn meniscus, and 101 of 129 cases were correctly classified as having a torn ACL. This study demonstrated that statistical methods applied to medical historical data can make a differential clinical diagnosis of an unknown knee joint problem with high degree of accuracy and with statistical significance. In the future, computerized medical diagnostic instruments can be constructed using these statistical methods.