Table 1 Preparation and execution of latent tree analysis
Stepa | Procedure | |
|---|---|---|
(i) | Statistical pattern discovery | Build three independent global latent tree models on the Lantern software Choose the model with the best BIC score for subsequent steps Obtain probabilistic co-occurring clinical features from each latent variable |
(ii) | Statistical pattern interpretation | Examine the quantitative relationships between latent variables and constituting clinical features by checking relevant probability distributions on Lantern Determine the TCM diagnostic pattern connotations for the latent variables from clinical perspective with TCM expertise Generate a list of potential TCM diagnostic patterns |
(iii) | Traditional Chinese Medicine diagnostic pattern identification | Based on TCM expertise, select only the potential TCM diagnostic patterns that contain all essential clinical features for subsequent steps Discard those that do not contain all essential clinical features |
(iv) | Traditional Chinese Medicine diagnostic pattern quantification | Construct a local latent tree model for each selected TCM diagnostic pattern on Lantern |
(v) | Traditional Chinese Medicine pattern differentiation rule derivation | Apply the local latent tree models to classify the participants Assign a soft label to each participant based on the probability of belonging to each TCM diagnostic pattern Derive score-based differentiation rules using the Naïve Bayes approachb |