| BackgroundCorona Virus Disease 2019(COVID-19)has the characteristics of fast transmission,rapid onset,and severe illness,which has long been a concern in the medical field and urgently needs effective and safe treatment measures.Traditional Chinese medicine(TCM)plays an essential role in treating COVID-19.Some TCM treatments have been proven to effectively relieve symptoms such as fever and reduce the rate of turning severe.However,there still needs to be overall effectiveness and safety evaluation of TCM for COVID-19 and personalized guidance for clinical doctors in selecting and applying different TCM treatments in complex and diverse real-world patient conditions.PurposeTo establish a "Research Database of Integrated Traditional Chinese and Western Medicine Treatment for COVID-19" based on electronic medical records and similarity matching algorithms.To conduct evaluations on the effectiveness and safety of generalized and commonly used TCM treatment measures through real-world research.To provide references for the personalized application of TCM treatment measures by establishing counterfactual prediction models.MethodsStudy 1:Methodology for constructing research database based on existing medical dataThere are still many challenges in building research databases due to the multi-source and complexity of real-world data.The construction of research databases can be divided into two stages:(1)determining the variable list and database architecture;(2)data governance.Determining the variable list and database architecture specifically which includes determining the variable list,evaluating existing data,clarifying data structure,and developing variable dictionaries.Data governance refers explicitly to data linking,extraction,verification,and cleaning.Study 2:Exploration of TCM prescription recognition based on real-world dataThe classification of real-world herbal prescriptions is achieved by creating a similaritymatching algorithm.The main steps include the predetermination of target prescriptions,drug importance labeling,preprocessing of prescriptions to be discriminated,matching and similarity calculation,determination of discrimination criteria,and removing duplicate prescription names.Study 3:Evaluation on the overall effectiveness of TCM in the treatment of COVID-19A multicenter retrospective cohort study was conducted based on a database.Using cumulative treatment of TCM for≥3 days during hospitalization as an exposure factor,patients were divided into TCM group and control group.Confounders were corrected by the inverse probability weighting method,and the deterioration rate and all-cause mortality were statistically analyzed by using a multiple Cox regression model.The deterioration rate and allcause mortality on days 7,14,and 28 were statistically analyzed by using a logistic regression model.Sensitivity analyses were conducted on the missing data cohort and the propensity score matching method cohort to evaluate the robustness of the results.The differences in the effectiveness of TCM in each subgroup were further explored by subgroup analysis.Study 4:Evaluation of the curative effectiveness of commonly used TCM treatment on COVID-19Based on the database,TCM treatment measures with≥150 users were screened,and effectiveness evaluation studies were conducted for each treatment.Exposure in each cohort was defined as receiving the treatment measure for a cumulative period of≥3 days during hospitalization.Patients were divided into exposure group and control group.The confounders were corrected by inverse probability weighting method and deterioration rates were statistically analyzed by using a multifactorial Cox regression model.Sensitivity analyses were performed for different statistical analysis models(semi-covariate multifactor Cox regression analysis and univariate Cox regression analysis),different definitions of exposure factors(cumulative time≥5 days and cumulative time≥1 day),and data sets under different definitions(excluding patients who used other treatment measures)to assess the robustness of the results.Differences in the effectiveness of selected interventions across subgroups were further explored by subgroup analysis.Study 5:A predictive model of TCM medication scheme for COVID-19 infectionBased on the database,27 variables including demographic information,disease conditions at admission,underlying diseases,vital signs,laboratory test results,and TCM treatment during hospitalization,were selected as predictors.After normalizing the data,a prediction model was constructed by using a neural network algorithm with a deterioration rate as the outcome.The model was internally validated by using 5-fold cross-validation,and its predictive ability was evaluated using the area under the receiver operating characteristic curve(AUROC),sensitivity,specificity,precision,accuracy,and F1 value.ResultsStudy 1:Methodology for constructing research database based on existing medical dataWhen determining the variable list and database architecture,first deconstruct the clinical problem to identify the critical variables needed for the study.Second,clarify the source and reliability of the key variable data,then select an appropriate data structure to store the data and establish a dataset.Third,record each variable’s information in the database’s variable dictionary in order to manage numerous variables.As for regarding data governance,the core of data linkage is to determine a unique and non-repeating patient ID sequence.The key of data extraction is homogenization so as to reduce differences between different personnel.Data verification can be achieved by combining sampling and comprehensive warranty to reduce workload while ensuring effective verification.Data cleaning focuses on developing appropriate cleaning rules for various problematic data,such as duplicate data,contradictory data,extreme values,and missing values identified during verification.Study 2:Exploration of TCM prescription recognition based on real-world dataA total of 78 target prescriptions for the treatment of COVID-19 were identified through the screening.Based on the similarity matching algorithm,7917(87.49%)of the 9049 prescriptions in the database could be identified.The algorithm can classify herbal prescriptions and provide a data basis for subsequent research.However,there are certain limitations in the algorithm’s consideration of the impact of herbal dosage on results,weight values,and determination criteria,which still require further research.Study 3:Evaluation on the overall effectiveness of TCM in the treatment of COVID-19A total of 2272 COVID-19 patients were included,with 1686 in the TCM group and 586 in the control group.After adjusting for confounding factors by using inverse probability weighting,the HR value of the deterioration rate in the TCM group was 0.52(95%CI:0.41 to 0.64,p<0.001)comparing with the control group.This result was consistent among patients with different severity levels at admission,and two sensitivity analyses confirmed the robustness of the results.In addition,the HR value of the all-cause mortality rate in the TCM group was 0.29(95%CI:0.19 to 0.44,p<0.001).The logistic regression analysis showed that the deterioration rate(OR=0.51/0.66/0.71)and all-cause mortality rate(OR=0.09/0.14/0.23)in the TCM group were lower than those in the control group on days 7,14 and 28.Subgroup analysis shows that TCM has a reduced effect on improving deterioration rates in patients under 40 years old and those with chronic respiratory diseases.Regarding safety,the proportion of patients with liver and kidney function abnormalities in the TCM group was lower than that in the control group.Study 4:Evaluation of the curative effectiveness of commonly used TCM treatment on COVID-19Eight common TCM treatment measures were selected.In the primary analysis,the Qi Deficiency of Lung-Spleen formula(HR=0.08),Banxia Houpo decoction(HR=0.18),Erchen decoction(HR=0.24),Maxing Shigan decoction(HR=0.36),Sijunzi decoction(HR=0.39),and Xiaochaihu decoction(HR=0.31)all reduced the deterioration rate of CO VID-19(p<0.001).At the same time,Lianhua Qingwen capsules(HR=0.97,p=0.762)and Xiyanping injection(HR=1.11,p=0.654)had no statistically significant effect on the deterioration rate of patients.In the sensitivity analysis of different statistical models,except for the Xiyanping injection,which had a higher HR value in the single-factor Cox analysis than the primary analysis results,the HR values of other treatment measures and statistical models did not change significantly.In the sensitivity analysis of different exposure factor definitions,except for Lianhua Qingwen capsules,the HR values of other treatment measures decreased in the sensitivity analysis with cumulative time≥5 days,with the HR value of Xiyanping injection decreasing to 0.42.In the sensitivity analysis of different defined datasets,except for the Sijunzi decoction,whose HR value increased to 0.50,the HR values of other treatment measures did not significantly increase.In subgroup analysis,Lianhua Qingwen capsules are more effective for patients over 60 years old or with severe COVID-19,and Xiyanping injection is more effective for patients with nausea and vomiting symptoms.Study 5:A predictive model of TCM medication scheme for COVID-19 infectionThe prediction model was constructed based on the neural network algorithm.Internal validation was conducted through 5-fold cross-validation,and the model had a good predictive performance with AUROC values of 0.995 in the training set and 1.000 in the validation set.Three simulated COVID-19 patients with different severity levels and their data were generated by computer.This model can calculate the probability of disease progression for each patient after receiving 8 different TCM treatment regimens,recommend the most suitable and appropriate TCM medication strategy,and provide a reference for clinicians to individualize the precise medication.Conclusion(1)The method of established a "Research Database of Integrated Traditional Chinese and Western Medicine Treatment for COVID-19" by identifying a list of variables,evaluating existing data,specifying the structure of the data set,specifying a dictionary of variables,data linking,data extraction,data verification and data cleaning,and using a similarity matching algorithm to identify TCM herbal prescriptions was verified.(2)According to the result of the retrospective cohort study,it confirmed that a full course treatment of TCM combing with conventional therapy may reduce the rate of deterioration and all-cause mortality in COVID-19 patients and it is generally safe.(3)The results of retrospective cohort study respectively confirmed the effectiveness of the Qi Deficiency of Lung-spleen formula,Banxia Houpo decoction,Erchen decoction,Maxing Shigan decoction,Sijunzi decoction,and Xiaochaihu decoction on reducing the deterioration rate of COVID-19,but further evaluation by randomized controlled trials are still needed.This study cannot determine the effectiveness of the Lianhua Qingwen capsule and Xiyanping injection on the deterioration rate,possibly due to the non-standard use of patent TCM medicine and underestimation of the therapeutic effect in the exposure group.(4)TCM treatment measures as predictors are innovatively used.Meanwhile,constructed counterfactual prediction models based on neural network algorithms.This action can provide a reference for clinical individualized medication by predicting the effectiveness of different medication regimens for various patients.However,it is necessary to establish a research-based database with larger sample size especially including TCM variables in the future.TCM recision medicine will be achieved through a predictive model that is more in line with the characteristics of TCM. |
PDF Full Text Request