Research On The Pattern For Electronic Medical Record Data-Driven Clinical Decision-Making

Objective:Data has become a basic strategic resource and a new production factor.Big data in healthcare is becoming a key driving force of medical decision support.Electronic medical record data is the core source of healthcare big data and a necessary part of real-world data.It plays an important role in the realization of the value of electronic medical record data and the optimization of clinical decision making to promote clinical decision-making driven by electronic medical record data and to give full play to the clinical decision support function of electronic medical records.However,there is currently no systematic pattern for how electronic medical record data drives clinical decision-making.Specifically,the analysis of important data types,characteristics and driving rules in data-driven decision-making is not deep enough,and there is a lack of targeted analysis of the driving process and influencing factors of electronic medical record data.Moreover,elements other than data in driving are neglected,especially the needs as a human being,which making it difficult to form an effective association and collaboration with users in actual scenarios.The research of the data analysis and transformation law in the clinical decision support process is all the more lacking,and a theoretical system has not yet been formed.Based on this,the study was oriented to clinical decision support,centered on the perspective of medical informatics,guided by the process theory and variable theory,information chain,Hall three-dimensional structure and other theories.It analyzed the driving constituent elements and focused on the analysis of the data characteristics of electronic medical records.constructed a pattern framework.The situational demands were determined as the driving goal,and the driving process model and influencing factor model were determined as the core.The “LDA+ template analysis” method was proposed to deeply explore the situational demands of doctors.The process of clinical decision-making driven by electronic medical record was analyzed,and the driving process model was constructed.The influential factors and mechanisms of electronic medical record data-driven clinical decision-making were analyzed,the influencing factor model was constructed.The research system of data-driven decision making and clinical decision support was enriched,the value of electronic medical record was promoted,and clinical decision support services were optimized.Methods:(1)The literature survey method and bibliometric method were used to analyze the status quo of relevant research,and clarify the research purpose and core issues.This paper used the literature research method and comparative analysis method to sort out the advantages and value of electronic medical record data-driven clinical decisionmaking,to analyze the driving constituent elements,and construct a model framework with situational demands as the driving goal,and driving process model and influencing factor model as the driving core.(2)According to the situational demands in the model framework,the “LDA+template analysis” method was proposed for the theme analysis of demands through the data collection of semi-structured interviews with clinicians.The Latent Dirichlet Allocation was used for preliminary demand identification.The identification results were used as the prior theme,and the interview data were returned for template analysis to understand the topic pattern of demand deeply.A fine-grained demand topic template was generated,and a highly structured demands hierarchy framework was constructed.(3)The model construction method was adopted for the driving process model in the model framework,based on the Hall three-dimensional structure,and the demand theme was used to construct the driving process theoretical model.The construction of the lung cancer database was empirically analyzed by case analysis,and then the Decision Tree,Logistic Regression,Random Forest,Extreme Gradient Boosting(XGBoost),Gradient Boosting Decision Tree,and Light Gradient Boosting Machine in machine learning were used to construct the electronic medical record data-driven lung cancer metastasis diagnosis model.(4)According to the influencing factor model in the model framework,a questionnaire was designed according to the constructed conceptual model of influencing factors for investigation,data were collected,the influencing factor variables of electronic medical record data-driven clinical decision-making were measured,the reliability and validity tests were carried out by factor analysis.The structural equation method was used to verify the influencing factor model and determine the influencing factors and influencing paths.Results:(1)The pattern framework was constructed,and the structure of the pattern was determined.Electronic medical record data-driven clinical decision-making also had its unique advantages such as relatively real-time and accurate data,relatively universal driving results,and easy integration,on the basis of the value of speeding up diagnosis and treatment innovation,improving clinical work efficiency and reducing medical risks.Its core components included human beings,electronic medical record data,technology,clinical decision information,and organization and environment.The collaborative components built the pattern framework for electronic medical record data-driven clinical decision-making,consisting of a driving target layer,a driving implementation layer,and a driving support layer.The framework took the situational demands as the driving target,and the driving process model and influencing factor model as the core.The driving model was human-centered,systematic,feedback-based,and interdisciplinary.(2)The demands hierarchy model was constructed and the driving goal was refined.Interviews were conducted with 17 clinicians to collect data for a demand analysis of electronic medical record data-driven clinical decision-making.Audio with a cumulative duration of about 727 minutes and normalized transcribed documents totaling more than 210,000 words were obtained.Through the “LDA+ template analysis” method,7 first-level demand themes,24 second-level themes,53 third-level themes,43 fourth-level themes,and 2 fifth-level themes were distilled under the view of the information chain.Taking the information chain as the core skeleton,the demands hierarchy framework was constructed,and the demand topics were classified into four levels.In addition,participants perceived risks in five aspects of data,technology,human beings,clinical decision-making information and organizational environment,such as unclear data authentication,low maturity of technology,and the possibility that doctors may be overly reliant on data and technology.(3)The driving process model was constructed,and the driving process was clarified.The driving process model was divided into goal dimension,logical dimension and knowledge dimension.The goal dimension was divided into seven demand themes,such as the intelligent recording of medical records,organization and extraction of clinical key information,and described the goal and behavioral orientation of the driving process.The knowledge dimension integrated multidisciplinary theory,method,technology,and expert knowledge.It described the subjects and professional knowledge supporting the goal and the logical plane from the perspective of disciplinary knowledge input.The logical dimension was divided into four steps,electronic medical record data generation and acquisition,data preprocessing and storage,modeling and analysis,and visualization and interpretation.It described the steps of electronic medical record data generation to produce clinical decision intelligence and solve demand goals.An electronic medical record data-driven lung cancer metastasis diagnosis model was empirically constructed and visualized.Among them,the prediction model based on the XGBoost algorithm performed better,with an accuracy of 0.81.(4)The influencing factor model was constructed and the driving influence mechanism was explored.Among clinicians from 24 provinces,municipalities,and autonomous regions,a questionnaire survey was conducted,and 309 valid questionnaire data were collected.The Cronbach coefficient of the questionnaire was 0.962.It was confirmed that the reliability and validity of the questionnaire were ideal through the exploratory factor analysis and confirmatory factor analysis.The influence path model of electronic medical record data-driven clinical decision-making was constructed based on structural equation modeling.The model explained 67% of the variance in willingness to adoption behavior.Organizational environment,performance expectancy,and data quality significantly and positively influenced the behavioral intention to adopt.Technology quality and data quality significantly and positively influenced performance expectancy.Technology quality,resistance to change,and perceived risk did not significantly influence behavioral intention to adopt.Conclusions:(1)The driving target layer of the pattern framework was the situational demands.The driving support layer took the element of the organizational environment as the guarantee,the element of human beings as the key,and the element of technology as the mean.Its driving implementation layer took the element of electronic medical record data as the logical starting point,took the element of clinical decision-making information as the driving product,and the driving process model and influencing factor model as the core.It highlighted the typical characteristics of electronic medical record data including time-dependence,highly unbalance,incomplete data,high-dimensional sparsity,real privacy,multi-source heterogeneity,and massive speed.The driving process model and the influencing factor model provided different but complementary perspectives for driving implementation regarding research questions,model description,empirical validation,practical application,and interpretation.(2)The demands hierarchy framework divided demands topics into four levels and7 categories of themes.There was a demand for intelligent recording of medical records at the data level.There was a demand for organization and extraction of critical clinical information for identification at the information level.There were 2 demands for disease risk prediction based on electronic medical record data and the refinement and supplementation of disease diagnostic and treatment experience and knowledge at the knowledge level.There were 3 demands for diagnostic assistance of disease,disease abnormality causes analysis,and assistance in formulating and recommending treatment plans at the intelligence level.In addition,the Latent Dirichlet Allocation modeling analysis method for mining Chinese interview data had the disadvantages of poor integration of correlation between words,easy to highlight the words in the interview outline,weak interpretability of the results,and rougher analysis.And it could be used in conjunction with template analysis.The setting of prior themes in the template analysis method provided an important embedding point for the combination of Latent Dirichlet Allocation modeling analysis.(3)The driving process was oriented to seven demand themes and carried out according to four steps: electronic medical record data generation and collection,data preprocessing and storage,modeling and analysis,visualization and interpretation under the support of the “trinity” of multidisciplinary knowledge and theories,methods and technologies,and expert expertise such as information resource management,data science,computer science,clinical medicine,and medical informatics.Overall,the electronic medical record data was pulled by the demand and supported by the cycle of knowledge resource flow,which was continuously undergoing transformational processing,generating clinical decision information and driving the whole driving process to operate towards the goal of realizing the clinician’s situational demands.(4)The influencing factor model revealed the influencing mechanism of performance expectancy,data quality,organizational environment,and technology quality on the adoption and application of electronic medical record data-driven clinical decision-making.Among them,the organizational environment had the greatest impact.Recommendations to promote the application and adoption of electronic medical record data-driven clinical decision-making were presented,which provided references for electronic medical record data empowerment and clinical decision support.