Data Mining Study On Effectiveness Of Acupuncture For Neck Pain Caused By Cervical Spondylosis Based On CER

Object iveIn order to solve the common demerits of data inconsistency and limitation of sample size in acupuncture clinical researches, the data mining methods was introduced under the strategy of comparative effectiveness research (CER) in this study as a new analytic approach for the clinical data of acupuncture as treatment for neck pain caused by cervical spondylosis. A treatment effect evaluation model was built based on a training data set origninating from a large clinical data sample. And after data training, the data mining model was used to differentiate the efficacy of acupuncture treatment in a small data set of another clinical trail. The objective is to explore the key data mining technology for CER research on acupuncture.MethodsThe research applied a combined research strategy of a randomized control trial (RCT) and data mining computer experiment. At first, a multicenter RCT was conducted to assess the effect of an optimized acupuncture treatment combined with intradermal needling. An acupoint shallow needling group and a non-acupoint placebo acupuncture group were set up as controls. The Northwick Park Neck Pain Questionnaire (NPQ) was used as primary outcome for treatment effect. The McGill Pain Questionnaire (MPQ) and the Short Form (36) Health Survey (SF-36) are applied as secondary outcome measures. The evaluation was performed from four aspects, i.e. evaluation based on western medical diagonosis, evaluation based on syndrome classification of Chinese medicine, evaluation based on morbid course, and evaluation based on pain intensity, thus the comprehensive effectiveness of the optimized treatment can be concluded. The SPSS version17.0software was used for data analysis for conventional statistic methods. The interntion to treat (ITT) principle was followed in our data processing, which included analysis of the Pro-protocol set (PPS) and the full analytic set (FAS) for interventional effect, analysis for blinding effect, safety analysis, compliance analysis and analysis for drop-out cases. In the computer experiment of data mining, the artificial neural network model and decision tree model were applied to the analysis of the data set acquired from the previous RCT. A comprehensive parameter of over patient-reported outcome (OPROO) was constructed as the primary efficacy variable for efficacy assessment. The detailed tasks included the application of the Kernel-as-similarity algorithm for correlations of different influencial factors, the application of Kernel Decision Tree algorithm to build the model for effect evaluation, the application of Local Learning algorithm to build a local model for effect evaluation to measure the similarity of different observed cases by their Euclidean distance, and the application of k Nearest Neighbour (kNN) algorithm to train the local set for later selection of signif ical representative cases. After the learning machine was trained by the data set of the above large sample RCT, a new small data set from another small sample RCT on acupuncture for neck pain caused by cerical spondylosis (CS neck pain) was introduced to the trained data mining model for efficacy judgement. The analytic strategy of conventional RCT and data mining was compared, and the feasibility and practicality of data mining technology for CER studies was explored in the study.ResultsThe multicenter RCT has totally recruited896patients when it was finished. There were103cases dropped out during the trial.793cases finished the intervention and follow-up survey. Thus the total protocol compliance rate was88.5%in this trial. In the effect evaluation based on Chinese medical syndrome classification, the optimized acupuncture treatment combined with intradermal needling had superior effect for the patients belonging to the Wind, Cold and Dampness syndrome and the Qi Insufficiency combined with Blood Stagnation syndrome. In the effect evaluation based on western diagnosis, the optimized treatment had superior effect for the patients diagnosed with Cervical Spondylosis (ICD-10code:M47.8) and Cervical Spondylosis with Radiculopathy (ICD-10code:M47.2). In the effect evaluation based on disease history, the optimized treatment had superior effect for the patients with a morbid course not longer than3years. In the effect evaluation based on neck pain intensity, the optimized treatment had superior effect for the patients with intermediate (VAS score from5to7) pain and severe pain (VAS score over7) compared to those with mild pain (VAS score from3to5). In the evaluation for efficacy, the patients were evaluated respectively in the end of their intervention, after one month follow-up, and after three months follow-up. The clinical effective rates in the endpoint of the intervention were90.4%in the treatment group,78.7%in the shallow needling group and67.5%in the placebo group (x2=40.995, P<0.001); the clinical effective rates after one month follow-up were87.7%in the treatment group,73.5%in the shallow needling group and64.5%in the placebo group (x2=38.306, P<0.001); the clinical effective rates after one month follow-up were85.4%in the treatment group,70.1%in the shallow needling group and63.4%in the the placebo group (x2=38.306, P<0.001). Therefore, we can conclude that the short, middle and long term clinical effective rates of the optimized acupuncture treatment were superior to the shallow needling group and the placebo group. In the evaluation of secondary outcomes, the score of McGill Pain Questionnaire declined in all observation points after the intervention and the follow-up survey. The improvement of MPQ score in the optimized acupuncture group was non-inferior to the shallow needling group or the placebo group with inter-group statistical significance. In the assessment of quality of life with SF-36, the patients in the optimized acupuncture group had superior improvement in all eight domains (i.e. Physical Functioning, Role Physical, Body Pain, General Health, Vitality, Social Functioning, Role Emotion, Mental Health), and they have superior improvement in four domains (i. e. Role Physical, Body Pain, Social Functioning, Role Emotion) of quality of life compared to the shallow needling group. Therefore, we can conclude the optimized acupuncture group has superior treatment effect over the shallow needling group and the placebo group in both short term and long term.The computer experiment of data mining indicated that the judgement ability of data mining model was improved after the local learning algorithm was applied. And the judgement accuracy was not significantly affected by the size of the learning set. The experiemtn of the Kernel Decision Tree model showed its accuracy was not affected by the size of the learning set after the data were preprocessed by Non-Dominant Sort algorithm, but it was correlated with the consistency of the data. In the experiment of the Kernel-as-Similarity algorithm, the judgement accuracy of the Kernal Decision Tree model reached72.45%±3.47%after the Kernel-as-Similarity algorithm was applied at a sampling rate of30%, and the its judgement accuracy was superior to other Kernel Decision Tree models with the sampling rate ranging from10%to90%. When the amount of learning machine reached certain extend, the judgement accuracy of the data mining model remained and became stable, and it implied that the increase of judgement accuary is not depended on the amount of learning machine. In the data mining study with the164-case small data set, the judgement accuracy was increased from65%-75%to75%-80%when the data mining model was trained by the896-case large data set and later optimized by algorithms such as support vector machine (SVM) or kNN. Thus it reflects the data mining approach is able to accurately evaluate the efficacy of acupuncture for CS neck pain.ConclusionThe multicenter RCT of this study reveals the clinical effect characteristics of the optimized acupuncture treatment combined with intradermal needling:it has best treatment effect to patients diagnosed with cervical spondylosis or cervical spondylosis with radiculopathy; the Chinese medical syndrome should be Wind, Cold and Dampness or Qi Insufficiency combined with Blood Stagnation syndrome; the morbid course should be within3years; the pain intensity should be over intermediate (VAS score over5); the treatment frequency should be twice a week with an interval more the1day; and the regimen should be8to10treatment session within in4weeks.The computer experiment showed the key influencial factor to the judgement accuracy of the data mining model is not the sample size, but the overall quanlity of the clinical data (e.g. data integratity and preciseness). The data mining model can efficiently assess the effectiveness and charateristics of the full clinical data set if a high quality training set is available. After trained by data set originating from a massive sample database, the judgement capacity of the learning machine was significantly improved for efficacy judgement and differentiation when it was applied to small clinical data sets for evulation. Furthermore, it is feasible to apply data mining methods to perform interim analysis of large clinical study, and or predict the endpoint outcomes to some extend before the trial is finished. Therefore, the data mining method can improve the efficiency of clinical studies, lower the difficulty of protocol implementation and save the unnecessary cost of large sample clinical trials which is common for the efficacy evaluation for new interventions.This study initially proved the feasibility of applying CER based data mining methods to the clinical research of acupuncture. The application of CER is helpful to explore and evaluate the efficacy of Chinese medicine. As the parameters of the overall efficacy variable is adjustable in accordance with the clinical and research needs, the sub-group analysis and inter-study analysis are earliy implemented under the framework of data mining analysis, and thus it is hopeful to reveal the clinical secrets of Chinese medicine.