Predicting Thyrotoxicosis in Patients Using a Set of Routine Tests: Adding their Rate of Annual Time-Series Variations to Self-Organizing Map-Based Predictive Model Improves Diagnostic Accuracy.

Difficulties have been associated with accurately diagnosing patients with thyroid dysfunction (PTD); however, measuring thyroid hormone levels in all individuals is challenging. We successfully constructed a prediction model for PTD by adopting pattern recognition methods using a combination of six routine laboratory tests, and identified 21 new PTD using our screening method, which was executed at two health check-up centers. In the present study, we newly introduced time-series variations in routine tests as additional parameters in order to develop the model by eliminating the influence of individual differences in routine tests. We constructed self-organizing maps (SOM) using the time-series traceable data of 13 PTD and 45 healthy individuals. We then investigated the locations of 140 projected false positives in our previous study on SOM and found that the number of false positives markedly decreased, thereby demonstrating the progression of our new model.