| Objective:Postherpetic neuralgia(PHN)is a common neuropathic pain in pain department.The pain can last for a long time and is difficult to be cured,which seriously affects the patients’ quality of life of.Many PHN patients even have suicidal tendency because of the pain.Studies have shown that additional analgesic treatments such as selective nerve root block and spinal cord stimulation analgesia in the early stages of herpes zoster can reduce the incidence of PHN,the combination of traditional Chinese and modern medicine has better curative effect than the single mode of analgesia treatment.But additional pain intervention also means additional medical risks and financial burdens,not every patients with shingles need additional analgesic therapy to prevent PHN,young people with shingles are more likely to heal without any residual symptoms.Therefore,it is a reasonable clinical decision to identify which patients are at higher risk of PHN and to give targeted additional analgesic treatment to these patients.Previous studies have revealed risk factors associated with the occurance of PHN,such as advanced age and acute pain intensity.However,these studies only remain in the perspective of qualitative studies,and can not provide a clear conclusion for a patient with multiple risk factors.The purpose of this study is to upgrade the qualitative research to quantitative research and give a probability prediction.With the increasing development of artificial intelligence and big data,these tools can make our clinical practice more specific and precise,In this study,related factors of PHN occurrence were analyzed from the perspective of integrated traditional Chinese and western medicine,and the probability of PHN occurrence in patients with herpes zoster was predicted by machine learning method.Method:First,a retrospective study was conducted to investigate the risk factors for PHN in patients with herpes zoster and PHN by univariate analysis.Secondly,these risk factors are taken as the characteristic factors of machine learning to build a prediction model.Because predicting PHN in patients with herpes zoster is a diconomical problem,it is suitable for supervised learning in machine learning,traditional supervised learning algorithms include logistic regression,random forest and artificial neural network,etc.In this study,these three algorithms are used to build prediction models and the prediction effectiveness of these three models is compared to select the best prediction model.Then the prediction model based on previous case data was applied to the prediction of new cases to test its predictive power,the concept of constitution classification of traditional Chinese medicine was introduced in these new cases to test whether it could be used as an independent factor related to the occurrence of PHN,and it was included in the machine learning model to see whether it could improve the prediction effectiveness of the model.Finally,the previous data and new cases data were fused to test whether increasing the sample size would improve the prediction ability of the model.Results:A total of 706 patients were included in the early stage of this study,through univariate analysis,the risk factors associated with PHN were age,herpes location,NRS score,CCI score,antiviral therapy,and immunosuppression status.The data of these six related factors were incorporated into machine learning,and the prediction accuracy of the three algorithms were:logistic regression 0.77,random forest 0.83,and artificial neural network 0.81,the random forest algorithm also has the highest recall ratio(0.76),precision ratio(0.78)and AUC value(0.88).which can be considered to be the best performing algorithm among the three algorithms(P=0.01).Then 204 new cases of herpes zoster and PHN were analyzed by TCM constitution classification,and it was found that the occurrence of PHN was related to constitution classification,PHN is more likely to occur in patients with blood stasis and qi-deficiency.Then,the random forest algorithm was used to predict 111 patients who were newly diagnosed with herpes zoster whether or not they will develop PHN,the prediction accuracy reached 79.28%,with the addition of TCM constitution classification,the prediction accuracy of random forest model was improved to 86.49%(P=0.04).The importance ranking of these 7 risk factors were NRS(0.23),CCI score(0.23),TCM constitution(0.17),herpes position(0.16),age(0.11),immunosuppressed status(0.05),and antiviral treatment(0.05).Finally,204 newly included cases were fused with 706 previous cases,and the prediction accuracy was improved from 0.83 to 0.86(P=0.07).Analyzed and discussed:This study first dug out the risk factors related to PHN through retrospective analysis,and then put the related risk factors into the machine learning model to predict whether patients with herpes zoster develop PHN,the accuracy of the prediction reached about 80 percent.In addition,traditional Chinese medicine also plays a very important role in the treatment of PHN,but TCM treatment needs to be guided by TCM theories.There is no analysis of risk factors related to PHN from the perspective of traditional Chinese medicine currently,this study incorporated the concept of constitution classification in TCM and found that different constitutions had differences in the incidence of PHN.By incorporating the classification of TCM constitution as a feature into the machine learning model,the accuracy of the model has been improved.The innovative points of this study are as follows:1.It is the first time to use machine learning method to predict the probability of PHN in patients with herpes zoster.2.Introducing the concept of TCM constitution science as a related risk factor.The constitution of TCM reflects the combination of innate endowments and acquired environment,It is a holistic classification standard and has high guiding significance in TCM clinical practice.3.The concept of CCI score was introduced for the first time in the analysis of the risk factors associated with PHN.CCI score can be used as an indicator to evaluate the comorbidities,previous studies often focus on a single disease,such as diabetes,leukemia,etc.,which cannot be analyzed from the overall complications.4.In this study,it was believed that the closer herpes to the head,the more likely PHN was to occur because of the higher density of spinal neurons in the head.The virus can infect more neurons per unit area and is more destructive,making it more likely to produce PHN,this is just a conjecture and needs further study.The shortcomings of this study are as follows:1.The included sample size is not large enough,although the 910 samples included in this study meet the requirements of the sample size,the nature of th e study determines that the more samples included,the better,it’s not very persuasive compared to studies with tens of thousands of cases.2.Related factors included are not comprehensive enough.Due to the limited availability of data,some relevant factors that appeared in the literature review were not included in this study,which can be supplemented in future studies.3.This study is a single-center study,and data from multiple centers can better reflect the overall picture of this problem. |
PDF Full Text Request