Statistical validation of strategies for Zang-Fu single pattern differentiation.

OBJECTIVE: To propose and validate a method to aid traditional Chinese medicine (TCM) physicians in differentiation of Zang-Fu single patterns.
METHODS: The procedure started with data collection and search on a knowledge database. Candidate patterns were selected and ranked according to the relative amount of explained exam's manifestations. Diagnosis identification was performed on a list of diagnostic hypotheses. Validation was conducted with 96,600 simulations of manifestation profiles obtained from database. Statistical performance based on confusion matrices was assessed for individual methods including inspection, auscultation and olfaction, inquiry, and palpation. Combined methods (inspection+auscultation and olfaction, inspection+auscultation and olfaction+inquiry) and four methods (inspection+auscultation and olfaction+inquiry+palpation) were also tested.
RESULTS: The highest accuracy was obtained with the inquiry method (89.7%), followed by inspection (70.7%), auscultation and olfaction (59.9%), and palpation (56.1%). The same sequence was found for both sensitivity and negative predictive values. Specificity and positive predictive values were almost equal and high (>99%) among individual exam methods. The combination of all methods provided the highest accuracy (93.2%), sensitivity (86.5%), and negative predictive value (88.1%), while sustained high specificity (99.9%) and positive predictive value (99.8%). The four methods presented the higher performance compared to combination of two or three exam methods as well as all single exam methods.
CONCLUSION: The proposed strategies present statistical evidence of its diagnostic performance and can be used to aid TCM physicians in making single pattern diagnosis according to Zang-Fu theory.