Yinghui Jin, Xiangying Ren, Lina Yu, Yongbo Wang, Qiao Huang, Xuhui Li, Xiangge Ren, Jiaao Lou, , Kuang Gao, Mukun Chen, Wenbin Hu, Xiantao Zeng＊, Hongcai Shang

1 Center for Evidence-Based and Translational Medicine, Zhongnan Hospital of Wuhan University, Wuhan, China.

2 College of Nursing and Health, Henan University, Kaifeng, Henan, China.

3 Department of Urology, Zhongnan Hospital of Wuhan University, Wuhan, China.

4 Henan University of Chinese Medicine, Zhengzhou, Henan, China.

5 College of Medicine, Wuhan University of Science and Technology, Wuhan, China.

6 School of Computer Science, Wuhan University, Wuhan, China.

7 Key Laboratory of Chinese Internal Medicine of Ministry of Education and Beijing, Dongzhimen Hospital, Beijing University of Chinese Medicine, Beijing, China.

Abstract The translation and implementation of clinical practice guidelines (CPGs) for Traditional Chinese Medicine (TCM) and Integrated Traditional Chinese and Western medicine is crucial to the adoption of medical science and technology, but the low operability and slow update of integrated traditional Chinese and Western Medicine guidelines, and the lack of integration between guidelines and clinical practice, result in the guidelines not having the desired clinical effects in practice.The application of Artificial Intelligence (AI) to the field of CPGs development aims to shorten the development time, optimize and accelerate the whole process of CPG’s development.This article summarized the current research and application status of AI in development and implementation CPGs for TCM and Integrated Traditional Chinese and Western medicine and proposed the method of Combining real world data and AI technology to enrich for TCM and Integrated Traditional Chinese and Western medicine.

Keywords:Artificial intelligence, Traditional Chinese medicine, Clinical practice guidelines

Background

Clinical practice guidelines (CPGs) are an important component of medicine, offering health care professionals recommendations for the optimization of patient care.CPGs are defined as “statements that include recommendations intended to optimize patient care that are informed by a systematic review of evidence and an assessment of the risks and benefits of alternative care options” [1].

Traditional Chinese Medicine (TCM) has more than 2,000 years of history and has gained widespread clinical applications [2].TCM and Integrated Traditional Chinese and Western Medicine have played a unique role in the prevention and treatment of diseases, such as SARS, influenza A, tumors and cardiovascular and cerebrovascular diseases[3].Development, dissemination and implementation of CPGs for TCM and Integrated Traditional Chinese and Western Medicine is a viable way for the internationalization of TCM and is also crucial to the adoption of science and technology of TCM for patient’s wellbeing [3-5].

But slow update and the low operability of CPGs based on TCM and Integrated Traditional Chinese and Western Medicine, and the lack of integration between CPGs and clinical practice, result in the guidelines not having the desired clinical effects in practice[3, 4, 6, 7].There is an urgent need to promote the rapid development and update of guideline recommendations so enabling them to be quickly and conveniently acquired and adopted by frontline clinical personnel to benefit the target population.

In recent years, Artificial Intelligence (AI) has demonstrated great progress in detection, diagnosis, and treatment of diseases [8-10].The AI guideline released by the Ministry of Science and Technology of the People’s Republic of China combined with four other Ministries also suggested increasing support for the application of AI in healthcare [11].In recent years there has been great progress in the application of AI to the field of TCM [12-14], and it has illuminated new ways of innovating CPGs for TCM and Integrated Traditional Chinese and Western Medicine.In this paper, by analyzing the application status of AI in these guidelines, we discuss and provide a perspective on how to shorten the formulation of CPGs, improve operability, innovate their dissemination and implementation mode through the use of AI, and puts forward some thoughts and suggestions for promoting the effective combination of AI and CPGs of TCM and Integrated Traditional Chinese and Western Medicine CPGs.

Application of AI to the development of TCM and Integrated Traditional Chinese and Western Medicine CPGs

The process of developing CPGs is a complex, timeconsuming, costly and multidisciplinary teamwork undertaking, no matter for TCM or Western CPGs, which usually takes 1-2 years to complete.The application of AI technology to the field of guideline development is conducive to improving work efficiency and reducing labor intensity [15].In addition, the use of AI technology makes possible the application of “l(fā)iving systematic review” [16] and the dynamic updating of guidelines [17].Based on AI technologies (such as machine learning, data mining, text mining and visualization), different groups of researchers have designed different software tools or platforms to simplify and improve the efficiency of the development process of CPGs.

Systematic review using AI-based tools

The application of AI to the field of CPGs development aims to shorten the development time, optimize and accelerate the whole process of CPG’s development.

In guideline development, a series of rigorous scientific steps are involved, among which the most arduous task is the systematic review of existing evidence.But now there is a turning point.A case study [18] has showed that using automated technology, a systematic review was able to be completed in only 2 weeks.This was thanks to the application of AI technology.Researchers at home and abroad are actively exploring tools for conducting systematic reviews.The evidence-based medicine AI-Reviewer, developed by Chinese researchers, is dedicated to assisting researchers in rapid screening of documents and is now being used in Peking University Third Hospital.In order to facilitate researchers in choosing appropriate automated tools, Marshell established a systematic review toolbox in 2015 [19], which is currently the most comprehensive online platform for collecting systematic review tools covering automated search, study selection, text analysis, data extraction and quality assessment.Table 1 lists several tools that use machine learning methods for systematic review, including, Abstrackr, ASReview, Colandr, Rayyan and RobotAnalyst.

In May 2016, Cochrane crowd, a researcher community platform for the text classification of RCTs was launched by the Cochrane Collaboration.The platform establishes a machine learning model, which can predict the possible usefulness of RCTs according to the topic and abstract, which is equivalent to an RCT Classifier.This model can exclude 60%-80% of unrelated studies and maintain more than 99% sensitivity [20].The Cochrane Crowd RCT model does not directly screen out the research that fully meets the set criteria, but after evaluating the title and summary of the relevant research, it integrates all the most likely RCT studies.This can then be followed by manual screening of the full text which narrows the scope of RCT research screening resulting in a reduced workload for later literature screening [20].

CPGs for TCM and Integrated Traditional Chinese and Western Medicine should actively apply literature screening, annotation, classification and data extraction methods or tools based on AI technology.At present, those methods or tools still need to be verified, and we did not find any studies which discussed or used AIbased methods to conduct systematic reviews for TCM.

The research has showed that the average update cycle of Integrated Chinese and Western Medicine guidelines is 7 years [3].When important new evidence appears, guidelines are not updated in a timely manner, and outdated recommendations may mislead clinical decision-makers.In addition, almost all the guidelines need to be updated by manual literature retrieval and evidence synthesis, and then entered the traditional guidelines writing and publishing process.Because this process takes a long time, when the guidelines are updated, it is likely that with the continuous publication of new literature, the updated guidelines already lag behind the latest research.So, tools realized the automated search, study selection and text analysis, which can be very helpful for update for systematic review and CPGs.

Beyond that, Yaolong Chen and Hongcai Shang have indicated that AI can be applied to screening for clinical problems, management of conflicts of interests, appraisal of CPGs and comparing recommendations from the different guidelines [21].For example, AI technology can quickly analyze the existing healthcare problems on a database or network, and automatically screen out the list of high-frequency problems, which can be used as an important reference for the CPGs to solve the problems.By using association analysis of information and data from different channels, AI can detect when an expert has attended the product promotion activities of an enterprise many times, or his immediate family members have a close relationship with the pharmaceutical enterprise, which can be used to substantiate the existence of conflict of interest by the conflict of interest management committee [21].

Guideline authoring and publication platform

MAGIC (Making GRADE an Irresistible Choice) is a non-profit foundation whose goal is to increase value in healthcare and reduce waste through an ecosystem of digital and trusted evidence [22].MAGIC is mainly composed of 3 key systems: Firstly, the Evidence Ecosystem: the ultimate goal of which is to increase value and reduce waste in health care and research by facilitating evidence flow.Secondly, BMJ Rapid Recommendations: The role of the rapid development system of guideline recommendations is mainly to form a complete methodological framework and system, enabling quick and high-quality production of guideline recommendations.Thirdly, MAGIC app: This is the core platform in the evidence ecosystem, allowing the creation, dissemination and dynamic update of digitized and structured evidence summaries, guide recommendations and decision-making aids [22].It is a web based collaborative tool that does not require any software installation and allows publication on all devices.The platform allows developers to conduct and publish guidelines and evidence summaries in a structured fashion.MAGIC group is exploring the possible uses of RDF and semantic web for guideline databases and its connection to other knowledge resources.

Guideline-based combined AI‐based CDSS

A systematic review in 2020 has shown that increasing the availability and dissemination of CPGs can promote their implementation[28].At present,the dissemination of CPGs is still limited to text form and clinicians are too busy to find and read guidelines resulting in lack of immediate acquisition of CPGs.In addition, CPGs developed for a single disease usually do not take into account multiple pathology, which would render them unsuitable for the complex and changing clinical situation.All these situations seriously hinder the effective implementation of CPGs in clinical decision-making practice [29-31].Only when the CPGs are digitized, intelligentized, and integrated into the clinical decision support system (CDSS) [32], can clinicians directly use these recommendations to assist themselves in clinical practice [33].

CDSS is an effective tool to promote the dissemination and implementation of CPGs, which are also in turn an important knowledge resources for CDSS [34].In the era of big data, AI technology can automatically learn from medical data, and extract the hidden rules or models, and then make intelligent decisions about the disease, which provides a new pathway for CDSS [35, 36].AI-based CDSS uses AI technologies such as reinforcement learning and deep learning to “l(fā)earn” from data such as guidelines, literature and medical records related to the nature of disease diagnosis and treatment, it can self-improve the knowledge base, rule base and decision engine model, and realize accurate and efficient intelligent comprehensive analysis and judgment [34, 37], and then provide advice for clinicians to consider for application in health monitoring, assessment and proactive intervention [38].

CDSS increasingly embodies the eagerly awaited application of AI and machine learning in patient care [39].For example, Watson For Oncology (WFO), which has been developed by IBM, is the most representative CDSS [38].It is composed of a reasoning engine, knowledge base and data extracted from clinical records [40].It can quickly read patients’ clinical medical records and retrieve published literature, guidelines and other relevant data to extract a series of treatment plans and suggestions [41].One study suggests that the use of clinical decision support and a variety of cognitive assistants will become routine in clinical practice within the next 25 years, and algorithms and AI technology will permeate almost every stage of decision-making [39].To achieve a higher-level CDSS, it is necessary to incorporate the following features: automated guidelines, algorithms based on guidelines, or algorithms that can generate the best care plan as fully as possible., CDSSs must also have the ability to manage big data including multiple complex forms of clinical records, disease registries, and patient surveys [42].AI models combined with evidencebased medicine evidence and high-quality clinical data will facilitate expert-level clinical decisionmaking tools across clinical settings [43].

A large number of clinical records, CPGs for TCM and Integrated Traditional Chinese and Western Medicine and evidence-based medicine research results are all important resources for TCM CDSS [44].Ontology technology, data mining technology, intelligent engine technology, machine learning technology and other relevant information technologies provide technical support for TCM diagnosis and treatment activities based on CDSS [45].TCM Clinical Decision System (TCMCDS),independently developed by the Information Institute of China Academy of TCM, is the one of the more mature CDSS [46].This system uses AI technology to construct the knowledge map of text data such as TCM clinical guidelines, experts’ experience, ancient and modern TCM endeavors to realize the decision support for TCM clinical diagnosis and treatment [47].

The development and application of important models and AI tools in the process of computer-interpretable CPGs

The first step to conduct guideline-based combined AIbased CDSS is changing the text-based CPGs into computer-interpretable CPGs.Table 2 shows the development and application of important models and AI tools in the process of generating computerinterpretable CPGs in recent years.

From Table 2, we can see that an increasing number of researchers have carried out more in-depth research in the development of models and tools in the process of computer-interpretable guideline by AI over the last 20 years.Researchers developed a variety of computerreadable CPGs expression models, such as the GLIF, GEM, SAGE, machine learning and CPG-RA models, which were based on the static text of CPGs and expressed in computer-recognizable language.Meanwhile, its related methods have been tried in a small range of guideline formulation, implementation and dissemination.However, the application of AI in the CPGs of TCM and integrated traditional Chinese and Western Medicine is still in its early stages.Digital and intelligent CPGs is an important method and trend to enable CPGs to be quickly and conveniently acquired and adopted by frontline clinical personnel to benefit the target population.However, the current research and use of digital or intelligent CPGs still heavily lag behind the development of the era of intelligence and information technology.

Table 1.Systematic reviews tools of functions and machine learning algorithms

Table 2.The development and application of important models and AI tools in the process of making computer-interpretable CPGs in recent years

Model/Tool Institution/Scholar Development Year Applications GuideLine Acquisition, Representation and Execution(GLARE) [60] Terenziani et al. 2008 A domain-independent system for acquiring, representing and executin CPGs A citation retrieval system composed of query expansion and citation ranking methods [61] Bui et al. 2015 Help to retrieve evidence for guideline development, and automatical find relevant citations for clinical guideline development The Unified Medical Language System (UMLS) [62] Becker et al. 2017 Provide a reference terminology and the semantic link for combinin clinical pathways to patient-specific information A framework for automated conflict detection and resolution in medical guidelines.Science of computer programming [63] Bowles et al. 2019 Present an automated formal framework to detect and resolve conflicts the treatments applied for patients with multi-morbidities focusing o medications A knowledge graph construction method based on clinical guidelines [64] Ziming Yi et al. 2020 Optimize knowledge, design and adjust the relationship structu between concepts, refine and improve the connection between example and use clinical guidelines for non-small cell lung cancer and corona heart disease as examples to establish knowledge graphs of two disease

With the rapid development of AI, knowledge graphs have become a research hotspot in the field of knowledge services.Knowledge graphs are a structured form to describe the concepts, entities and their relationships in the objective world.It expresses the information from the Internet in a form closer to the human cognitive world, and provides a better ability to organize and understand the huge amount of information on the internet.Knowledge graph brings vitality to internet semantic search [64].Tong Yu et al.constructed a large-scale TCM knowledge graph, which integrated terms, documents, databases and other knowledge resources.This knowledge graph can facilitate various knowledge services such as knowledge visualization, knowledge retrieval, and knowledge recommendation, and helped the sharing, interpretation, and utilization of TCM health care knowledge [12].In this research, the existing TCM databases were used to provide the data resources, although CPGs was not the only one or specified designed data resource, we still believe CPGs which have been developed using scientific and rigorous evidence-based medicine methodology should be the best data resources for TCM knowledge service.Knowledge graph represents a relatively new method for the systematic organization and deep analysis of the TCM health knowledge system, and have application prospects in knowledge management, service, education, and training in the field of TCM health care.

Combining real world data and AI technology to enrich TCM and Integrated Traditional Chinese and Western Medicine CPGs

There are still great deficiencies in the implementation of CPGs in clinical practice [4, 65].The low operability of the guidelines is the primary factor affecting implementation as their recommendations often cannot not solve the complicated patient problems in the clinical environment [66, 67].For example, a guideline is usually formulated for a single disease and is based on evidence from clinical studies focused on internal authenticity, but in the actual clinical environment, doctors are faced with complex clinical problems, such as multiple pathology, difficult or rare cases, advanced age, young children, or pregnancy and childbirth.These problems usually cannot be addressed based on the guideline’s recommendations.

With the development of medicine and information technology, real world data (RWD) has rapidly emerged in the fields of medical care, drug supervision, and medical insurance, and has become an indispensable part of the medical and health industries.RWD are data relating to patient health status and/or the delivery of health care, routinely collected from a variety of sources [68].It can be derived from a wide range of sources, such as hospital electronic medical records, residents’ electronic health records, traditional epidemiological studies (e.g., classical cohort studies), administrative databases (e.g., medical claims, death registers), surveillance (e.g., spontaneous adverse drug events monitoring), or personal devices (e.g., regular blood pressure measurements using mobile devices) [69].Different from the research environment of traditional clinical trials, RWD emphasizes that the data comes from the actual clinical medical environment, and the data generation and collection processes are consistent with the actual clinical medical practice (see Figure 1).

Figure 1.Thinking and methods of Combining real world data and AI technology to enrich TCM and Integrated Traditional Chinese and Western Medicine CPGs

Due to the wide range of sources, a large amount of RWD is generated every day in the actual clinical environment.In recent years, an increasing number of studies have used RWD to conduct real world research to generate real world evidence [70, 71], which could be used to develop CPGs.Meanwhile, with the development of AI technologies such as knowledge graphs and deep learning, there are also studies exploring the use of AI technology in the field of TCM [72-74], for example, by mining the clinical core prescription in real world data, developing the auxiliary algorithm of TCM syndrome differentiation and treatment named "intelligent analysis system of TCM prescription", the inheritance of famous TCM doctor's experience has been realized.A study has reported using electronic medical record data and PrTransH algorithms to construct medical knowledge graphs [75].But at present, to our knowledge, there is no study which explores the use of RWD and AI technologies to complement the guidelines during the process of using them in clinical practice.Constructing knowledge graphs as the carrier of knowledge of the guidelines, and then extracting information from the RWD using deep learning and other technologies to complement guideline-based knowledge graphs is a feasible way to solve the problem of low operability of CPGs.

Due to the large-scale and diverse nature of, RWD it is often messy, repetitive, and incomplete, for example, there are many repetitions of text information in clinical electronic medical records, which hinder its direct use.Therefore, before using RWD to enrich CPGs, it is necessary to use AI technologies to preprocess the data, such as using the SimHash algorithm to de-redundant the data.For the data set obtained after preprocessing, technologies such as recurrent neural network and Bootstrapping algorithm can be used to extract medical entities in the guidelines to form a RWD heterogeneous network.This heterogeneous network is then aligned with the knowledge carrier, such as the guideline-based knowledge graph which is constructed to compliment the guideline-based knowledge graph.Finally, technologies, such as TransE model and MetaPath2Vec, can be used to enhance the reasoning and knowledge mining of the knowledge graph, helping discover unclear implicit relationships between different areas of medical knowledge.

At present, using RWD and AI technology to complement the CPGs of integrated TCM and Western medicine is still a relatively new research direction.However, with the accumulation of RWD and the development of AI technology, we believe the realworld data should be extracted by techniques such as big data deep learning and data mining, so as to transform real-world data into real-world evidence, resulting in enrichment of the knowledge graph formed by the guidelines with real-world evidence and this exploration will eventually become mature and perfect, promoting the operability of CPGs for TCM and Integrated Traditional Chinese and Western medicine, and promoting the transformation of medical science and technology achievements into clinical practice.

Above all, we discussed how to use AI technology for development and implement, particularly in systematic review using AI-based tools, guidelinebased combined AI-based CDSS, models and tools of AI in the process of computer-interpretable CPGs, and combining real world data and AI technology to enrich CPGs (see Figure 2).

Figure 2.Thoughts on AI for the Development and Implementation Guidelines for Traditional TCM and Integrated Traditional Chinese and Western Medicine

Discussion:Challenges and Future Directions of AI in development and implementation of Guidelines for TCM/Integrated Traditional Chinese and Western Medicine

Ethical Implications

From an ethical and legal perspective, as more and more personal information and clinical data are used in AI research, the issues of data security and personal privacy are becoming increasingly prominent [76].Under the automatic diagnosis decision-making mode of AI, the attribution of doctor-patient responsibility also needs to be clearly regulated by relevant laws and regulations [77].

The black box nature

Challenges also arise from the nature of AI and the manner in which it is developed.Some computational reasoning methods in AI such as neural networks are considered black boxes to end-users, so machine learning algorithms could also be subject to biases [78].machine learning algorithms, particularly those based on artificial neural networks, make inscrutable predictions and for these algorithms it is usually harder to detect error or bias [79].

Perspective for development of AI in TCM and integrated Traditional Chinese and Western Medicine

First of all, further studies are badly needed for developing the wide ranging application of AI tools in the production of systematic reviews, and this is the most important priority research area in AI-based guideline development.

Secondly, Cooperation should be strengthened between guideline development committees and information technology experts, so that the TCM essence can be retained and enhanced by AI technology.Secondly, the key to the development of AI to TCM is to achieve the standardization of TCM diagnosis and treatment data, which can lead to AI classification and processing of TCM real world data.This in turn can provide decision support for TCM clinical diagnosis and treatment services.

AI-powered medical technologies are fastly evolving into feasible solutions for clinical practice.Despite a series of significant achievements in the application of AI in the field of healthcare, its application in the field of CPGs is still in its infancy.The 10th Asian-Pacific Conference on Evidence-Based Medicine proposes a trinity development model of "Evidence-based Medicine, Chinese Medicine and AI" to explore the cross-integration of evidence-based medicine, TCM and AI [80].In terms of the speed of scientific and technological progress and the depth of interdisciplinary cooperation, the combination of AI technology and guidelines is on the way.AI is likely to lead the new era of CPGs development.