Construction Of TCM Syndrome Classification Forecast Model For Diabetic Nephropathy And Evaluation Of Its Recognition Performance

Diabetic Nephropathy (DN) is a common complication of diabetes mellitus. Chinese medicine (CM), boasting of the characteristics of treatment based on pattern differentiation (TBPD), shows proven efficacy in treating DN with respect to its positive effect on kidney protection and disease progression. Attention has long been paid to syndrome because of its significance to TBPD. If CM is to be accepted internationally, evidence that provides strong base for CM theory shall be retrieved from epidemiological investigation and be elaborated in the language accountable for mainstream sciences.The current bottlenecks to research on syndromes include:1. The simultaneous presence of diversity and unity of syndrome classification and absence of consensus on a standardized pathway to practice it that complicate its process of standardization; 2, As a non-linear complex system, the reductionistic interpretation of CM is furthered complicated by its characteristic on mega-parameters and mega-combiniations; 3. The determination of syndrome, much alike a black box, is realized through the merger of information from a doctor’s experience and knowledge, natural or social, which is characterized by fuzziness; and 4. The significance of each symptom weigh differently to the identification of a syndrome, making the quantification and objectification almost impossible.The essence of syndrome-related issues is to study the classification of significant symptoms identified from "four diagnostic approaches". So I postulate the commonly used algorithms for classification in statistics can be applied in the research of CM syndrome to better address the bottlenecks. I practiced it in a pool of clinical data on renal insufficiency and literatures on DN to discover potential rules that guide further studies dealing with these bottlenecks.Method:1. Apply classification algorithms including discriminant analysis, artificial neural network, support vector machine and random forest in the diagnosis of CM syndromes to compare the modeled results with real-world results. The the performance of each model is evaluated to find an effective approach to deal with non-linear, high dimensional, fuzzy information in syndrome research and to assign proper weight to each element.2. Make a thorough literature search with strict inclusion criteria to retrieve DN studies with or without treatment according to disease phases. Evaluate the retrieved literatures with respect to their quality to obtain standardized data. Then a study based on symptom-based syndrome differentiation is performed. Specific approaches include comparison of symptom frequency, exploration of how disease features relate to symptom elements, data mining for discovering prescription rules in treating prevalent patterns based on association rule. Finally, based on findings, the rule of CM composition of DN prescriptions.Results:1. Among the models based on classification algorithm we have developed, the Fisher discriminant model presented prediction accuracy of 72.84%, nearest neighbor model of 81.48%, K-nearest neighbor model of 71.60%, feedforward neural network model of 75.31%, support vector machine model of 85.19% and the Random Forest model of 86.42%. The advantages of these models are the effective solution of Fisher model to non-linear problems in the research of TCM pattern, of artificial neural network model to both non-linear problems and complex merger with indistinguishable process, of support vector machine to deal with the complexity of syndrome data caused by the large number and high dimensions of data, and of random forest model to the prediction of pattern classification with objective weight assignment. Combination application of algorithms will maximize the performance to better solve problems facing pattern studies.2. The symptom-symptom multi-collinearity is one of the problems facing TCM research. To analyze the variables by principle component analysis and factor analysis before performing cluster analysis is an effective way to better address the problem. With this approach, we have reduced the dimensionality of TCM symptoms to 7-9 syndromes, justifying the TCM classification of diabetic nephropathy into qi deficiency, blood deficiency, yin deficiency, yang deficiency, phlegm dampness, dampness turbid and blood stasis. The joint usage of k-mean cluster and pedigree cluster realizes the visual presentation of distributive spectrum. This gives a clear and concise pattern distribution of diabetes nephropathy so that the conventional empirical classification of patterns can be avoided.3. The authors have identified 171 articles relevant to pattern classification and 208 articles relevant to phase-based pattern classification of diabetes nephropathy. These have proven the characteristics of diabetic nephropathy of simultaneous deficiency and excess. The primary syndromes presents as qi deficiency, blood deficiency, yin deficiency, yang deficiency, phlegm dampness, dampness turbid and blood stasis. Yin deficiency and blood deficiency presents through the course. In the early stage, the disease implicates kidney, liver and spleen, and as progressing, it further implicates spleen and kidney that may transferred to lung and heart. The final condition can be summed up as multiple viscera implication and 5 viscera impairment. This is an effective solution to the problem facing TCM of variability of pattern differentiation and pattern standardization.4. Considering the pattern element distribution, this study summarizes the rule of prescription of diabetic nephropathy, The results identified several commonly used pair of herbal medicines including yam and moutan, dogwood and moutan, and yam and pore.This shows that the major treatment principle of diabetic nephropahty is by "benefit qi and transform into yin" that mainly focus on benefiting liver and kidney yin coupled with removing blood stasis. The commonly adopted prescriptions in clinical practice include Jiuwei Dihuang Pill, Shenqi Pill and Shenqi Dihuang Decoction. In the early stage, the treatment principle is benefiting qi and nourishing yin with removing blood stasis. In the middle stage, the treatment usually put emphasis on benefiting spleen qi. with dogwood and peach as the most common medicines. This is to achieve the effect to benefiting qi and nourishing yin with removing blood stasis.Conclusion:1. Artificial neural network, support vector machine and random forest can be applied in the classification of symptoms to establish syndrome model, which is a long recognized bottleneck in CM. Among the approaches, random forest and support vector machine showed effective prediction performance with high exactness. Besides, random forest shows additional benefit by its function to weigh CM symptoms so that the one contributing most to DN can be found. The 2 approaches are recommended for future syndrome studies.2. We provide a sum-up standardized process for future syndrome studies:(1) Based on the guidance of pattern differentiation according to symptom element, reap the concise and core elements of the data. (2) Select proper approaches according to research purpose. Complement the insufficiency of single approach by introducing more than one approach. (3) Provide results with strong-evidence...