Research On Framework And System Of Clinical Data Quality Assessment

Clinical data contains a large amount of information about diseases,diagnosis,treatment,and healthcare,which is an important data resource for discovering clinical evidence and promoting medical innovation.However,the quality problems of clinical data have become a huge obstacle that hinders the effective utilization of clinical data.Developing clinical data quality models,objectively assessing the quality of clinical data,and discovering data quality problems are significant for improving data quality and facilitating effective utilization of clinical data.Conducting data quality assessment,the evaluator needs to define data quality assessment items according to data quality models for specific scenarios,and then perform corresponding assessment based on quality assessment items.However,although the existing models have included indicators,most of the indicators do not accurately describe the corresponding data quality problems and lack objectivity and operability.The definition of quality assessment items relies on the professional knowledge and subjective experience of the evaluator,which leads to the inconsistent of quality assessment items defined by different evaluators,and further causes inconsistent and biased assessment results.Meanwhile,the implementation of data quality assessment items lacks automated approaches.Although there are rule-based tools for data quality assessment,the definition of rules relies on manual approaches;Some items which can not be performed by rules,such as data provenance,are still performed manually in clinical practices.Due to the subjective experience or oversights of the evaluator,the implementation process also leads to inconsistent and biased assessment results.It is difficult to objectively describe and compare the quality of clinical data based on inconsistent and biased assessment results,which affects the continuous improvement of data quality and obstacles to the effective utilization of clinical data.This paper studies the clinical data quality model and automated quality assessment technology,which improves the objectivity and operability of the evaluation and reduces the deviation caused by the subjective experience or oversights of the evaluator,thereby improving the consistency of the clinical data quality assessment.The major works including:1.This study summarized the dimensions of data quality in related researches and integrated the knowledge of data quality from literature and data quality reports.Then,this study developed an indicator system that including 43 indicators of data quality assessment by decomposing,multiplexing,merging,and summarizing the integrated knowledge.The quality dimensions and indictor system consist of a new data quality model.Experts were invited to modify and refine the data quality model through group discussion.Finally,the results of an evaluation experiment showed the developed data quality model of this study is able to correspond to data quality assessment items from an actual clinical case more accurately compared to traditional models,which verified the objectiveness and operability of our data quality model.2.On the basis of analyzing the assessment object of each assessment indicator and the most suitable assessment technology,this study designs the formal expression paradigm of the data quality assessment item corresponding to each indicator.The assessment condition,assessment object,and quality constraints of each indicator are structuralized according to the most suitable technology when performing the indicator.And,a clinical data quality assessment technical framework is established,which demonstrates data quality indicators to automated implementation of specific assessment techniques.According to the comparison experiment of several national data quality standards,94% of the data quality evaluation items can be accurately expressed through the paradigm and are able to be automatically executed by the most suitable evaluation technology,which further improves the consistency of data quality assessment.3.Aiming at the problem that the rule definition in the rule-based data quality assessment tool still relies on manual labor,this study developed an automatic approach to create data quality rules automatically based on clinical information models.For the standardized clinical information model(the open EHR model is an example for implementation in this study),the quality constraints information of each data item contained in the model is automatically extracted by analyzing the expression file of the clinical information model.The corresponding rule template is constructed based on the indicator formal expression paradigm.Then,the extracted quality constraint information and corresponding rule template are automatically converted as computer-executable data quality assessment rules are,which improves the efficiency of the definition of data quality assessment rules while greatly reducing errors caused by subjective experience,thereby improving the consistency of clinical data quality evaluation.4.Due to the need to automatically verifying the traceability of clinical data,this paper proposes an automated data verification approach for multiple data sources such as paper-based clinical information and textual medical records.Firstly,the traditional optical character recognition technology is improved based on machine learning,which can automatically extract the clinical information of the paper-based clinical documents;secondly,named entity recognition technology was used to retrieve key information from textual medical records;and then the relationship between different data sources is analyzed to construct a verification model oriented to multiple data sources.Clinical practice shows that this method can automatically analyze and verify the consistency of the researched clinical data with corresponding paper-based or textual original data source.5.Based on the clinical data quality assessment technical framework,this paper designed and implemented a highly scalable clinical data quality automated assessment system,which can dynamically "plug in" the most appropriate automated evaluation module for different assessment indicators,including but not limited to the automatic rule creation module and multiple sources data verification module,in order to adapt different application requirements and technological levels.6.Based on the above methods and system,we established an implementation path of clinical data quality assessment for practical application scenarios,and implemented it on a coronary heart disease registry in our country.Compared with traditional methods,the evaluation items constructed by different clinical experts based on the data quality model of this paper have better consistency,and the application of automated evaluation technologies also further reduces the human bias of the assessment results,which can implement data quality assessment more comprehensively and efficiently,and finally promote the improvement of clinical data quality.