Datasets:

---

ExponentialScience

/

DLT-Scientific-Literature

like 0

Follow

Exponential Science 4

ERROR: type should be string, got "https://doi.org/10.1007/s10916 022 01892 2\n\n**REVIEW**\n\n# Automating Electronic Health Record Data Quality Assessment\n\n**[Obinwa Ozonze[1] · Philip J. Scott[2] · Adrian A. Hopgood[1]](http://orcid.org/0000-0001-5385-7954)**\n\nReceived: 10 September 2021 / Accepted: 15 November 2022\n© The Author(s) 2023\n\n**Abstract**\nInformation systems such as Electronic Health Record (EHR) systems are susceptible to data quality (DQ) issues. Given the\ngrowing importance of EHR data, there is an increasing demand for strategies and tools to help ensure that available data\nare fit for use. However, developing reliable data quality assessment (DQA) tools necessary for guiding and evaluating\nimprovement efforts has remained a fundamental challenge. This review examines the state of research on operationalising\nEHR DQA, mainly automated tooling, and highlights necessary considerations for future implementations. We reviewed\n1841 articles from PubMed, Web of Science, and Scopus published between 2011 and 2021. 23 DQA programs deployed in\nreal-world settings to assess EHR data quality (n = 14), and a few experimental prototypes (n = 9), were identified. Many of\nthese programs investigate completeness (n = 15) and value conformance (n = 12) quality dimensions and are backed by\nknowledge items gathered from domain experts (n = 9), literature reviews and existing DQ measurements (n = 3). A few DQA\nprograms also explore the feasibility of using data-driven techniques to assess EHR data quality automatically. Overall, the\nautomation of EHR DQA is gaining traction, but current efforts are fragmented and not backed by relevant theory. Existing\nprograms also vary in scope, type of data supported, and how measurements are sourced. There is a need to standardise\nprograms for assessing EHR data quality, as current evidence suggests their quality may be unknown.\n\n**Keywords Electronic health record (EHR) · Data quality · Data quality assessment · Automation**\n\n\n### Introduction\n\n#### Electronic health records (EHRs)\n\nElectronic health record (EHR) systems play an integral\nrole in today’s healthcare practice, enabling hospitals and\nother health organisations to consistently collect, organise,\nand provide ready access to health information. These health\ninformation systems have arguably become the standard\nfor modern healthcare practice and are increasingly being\n\n- Adrian A. Hopgood\nadrian.hopgood@port.ac.uk\n\nObinwa Ozonze\n\nobiozonze@gmail.com\n\nPhilip J. Scott\n\nphilip.scott@uwtsd.ac.uk\n\n1 School of Computing, University of Portsmouth,\nBuckingham Building, Lion Terrace, Portsmouth PO1 3HE,\nUK\n\n2 Institute of Management and Health, University of Wales\nTrinity Saint David, Lampeter SA48 7ED, UK\n\n\nadopted globally in many health organisations to enhance\ncare coordination and outcomes [1–3]. They are also typified for collecting massive amounts of health data that are\nmore reflective of the real world, with great potential for\ninvestigating a wide range of research at lower costs [4, 5].\nRecent studies also show growing efforts to aggregate EHR\ndata and, using artificial intelligence techniques, explore\nEHR datasets to develop models that can help improve\ndecision-making and accelerate medical innovations and\nother secondary use objectives [6]. Secondary use (or reuse)\nhere generally refers to non-direct care activities, including\neducation, medical innovations, quality monitoring, public\nhealth surveillance, budgeting, and other commercial activities [4, 7].\n\n#### EHR data quality\n\nThe growing reuse of EHR data for secondary use can also\nbe attributed to the expectation that it is a factual representation of patient conditions, treatment, and outcomes.\nThese facts could be in the form of patient demographics,\ndiagnoses, details of laboratory and pathology examinations,\n\nV l (0123456789)1 3\n\n\n-----\n\nprocedures performed, and medications ordered and administered records. Other types of documentation available in\nEHRs include admission and discharge summaries, lifestyle\ninformation and referral letters [3, 8]. Typically, healthcare\nproviders capture the above record types using EHR forms\nand templates, scanning and speech-to-text tools [9]. Data\nmay also be pulled into EHRs from other sources, including\nelectronic measuring tools, clinical systems, and external\ndata repositories.\nNevertheless, as with many other information systems,\nEHR data can be prone to variable levels of quality, particularly in terms of completeness, correctness, consistency,\nconformance, plausibility, and timeliness [10, 11], and are\nnot always ready for meaningful analysis without considerable preparatory work. For example, several studies report\nmissing timestamps and records, implausible data entries,\nvalues outside normal ranges, and duplicates. Figure 1 presents a taxonomy of commonly reported error types in EHR\ndata. Interested readers can see [10–14].\nData quality (DQ) problems, particularly in EHRs,\nare the by-product of many social and technical factors,\nincluding people-related factors like work pressures, indirect data capture, misinformation from patients and other\nsloppy documentation practices [15–17]. Other factors\nsuch as variations in clinical practice and the lack of standardised protocols for data collection, non-intuitive EHR\nsystem design, prolonged or unsuccessful deployments,\nand organisational factors like workflow disruptions, staff\nrotations, computing aptitude and inappropriate use, such\nas copy and paste cultures, also inadvertently encourage\nthe capture of low-quality health data [18–21].\n\n**Fig. 1 Examples of data quality problems in EHR data**\n\n## 1 3\n\n\nUnfortunately, the cost of data quality problems is usually high, especially in industries like healthcare, negatively impacting patient safety, the quality of practice,\nresource management, and the credibility of clinical studies. Today, many medical errors have data errors as their\nroot cause [22]. Data errors also affect care coordination\nand threaten operational efficiency, making it challenging\nto track programme success or respond to emerging threats\n\n[23, 24]. Equally, clinical studies and decision support\ntools based on EHR data also spend large sums of money\non data preparation and still risk producing misleading\noutcomes [25–28]. There is also the consequence of an\nincreasing volume of unusable EHR data. Given the critical impact of these DQ problems and the high propensity\nto reuse EHR data, measures to ensure that available EHR\ndata are suitable and appropriate for intended use cases\nare essential.\n\n#### EHR data quality management\n\nPrior studies note that ensuring that some given data are fit\nfor use broadly involves four main steps: definition, measurement, analysis, and improvement activities [29–31], as\nshown in Fig. 2. The first step: definition, generally focuses\non specifying the context of use, data elements of interest,\ndata problem or dimensions to investigate. Measurement is\nthe second step, and it is used to ascertain the DQ status of\nthe dataset. Usually, this involves identifying problems in\nthe given dataset and reporting the dataset’s status based\non earlier criteria. The outcome of the measurement step is\ntypically a collection of records with the data problems of\n\n\n-----\n\n**Fig. 2 Typical DQ assessment and management framework**\n\ninterest and metrics depicting the degree of the identified\ndata problems in the data sample. The third step, analysis,\nentails assessing the identified data problems and estimating their impact on the specified context or root causes. The\nmeasurement and analysis steps in the literature have come\nto be generally understood to mean assessment. The final\nstep comprises activities to improve or make the dataset\nmore fit for the intended use case, such as preventive and\ncorrective procedures.\nIn contrast to other steps, there is a considerable amount\nof research on understanding and defining EHR DQ: data\nerror dimensions, taxonomies, and quality indicators [10, 11,\n32]. Several studies also present preventive interventions for\nimproving EHR data collection and management processes.\nFor example, some studies advocate continuous training in\nthe use of EHR software, enforcement of standards to curb\nvariations in documentation practice, more focus on data elements commonly needed for secondary use, giving patients\nmore access to their data, and providing tangible incentives\nto encourage accurate documentation [13, 28, 33, 34]. Other\nstudies also advocate better usability in EHR design, such as\ntailoring workflows to match clinical processes, and intuitive interfaces and documentation support like tooltips and\ninput masks to guide users when in doubt and promote best\npractices [35, 36].\nNonetheless, assessing EHR data quality, necessary for\nroot cause investigations, documentation training, data\ncleansing works, and ascertaining if implemented preventive\n\n\nand corrective interventions yield positive results, has\nremained a challenge. In many cases, data errors are rarely\nreported or even recognised when they occur. According to\na clinical leader in one study, “…no one knows how bad data\nis in hospitals – on a good day, it is bad; on a bad day, it is\nterrible…” [37]. Meanwhile, a comprehensive data quality\nassessment (DQA) ensures that available EHR data are complete, consistent, and fit for use. This assessment is critical\nas the absence of evidence (quantitative) of the extent of\nthe DQ problems makes creating baselines for tracking and\nprioritising interventions challenging [38, 39]. In addition,\nthere are many potential benefits that EHR data consumers\ncan derive from DQA, including improving the efficiency\nof data collection tools, reducing the cost of preparing EHR\ndata for analysis, enabling clear interpretation of outcomes,\nand deepening the global knowledge of disease and treatments [22, 40].\n\n#### Study objectives\n\nSeveral methods for assessing EHR data quality have been\npublished in the last decade [10, 41–43]. However, many\norganisations implement them in an ad-hoc and manual\nmanner, primarily via in-person audits and desk reviews\nthat involve significant human reasoning and time, which are\nunsuitable for large datasets, time-constrained use cases, and\ntasks requiring repeated assessments [44–46]. In addition,\nthe outcomes of these ad-hoc assessments are not readily\nreproducible as they are often conducted inconsistently, with\nassessors having varying skills and background knowledge\n\n[47, 48].\nGiven the high propensity for reusing EHR data, there is,\ntherefore, a need for reliable and automated tools that can\nhelp assess EHR DQ consistently, estimate the impact of\nidentified errors, and manage any risks involved before use.\nThis requirement is even more crucial now, with the growing\ncalls for improved transparency and confidence in EHR data\nmanagement [10, 11, 22]. As with developing most complex\nsystems, an explicit understanding of necessary components\nand their intricacies is also essential.\nHence, this review examines the state of research on\nEHR DQ, particularly recent approaches employed by\norganisations and studies to develop or implement dedicated\ntooling for assessing EHR DQ. Our primary goal is to identify necessary features and considerations that could guide\nEHR DQA tooling, not limited to dimensions and assessment methods [10, 41, 49]. This work also seeks to extend\nCallahan et al. [50]’s study comparing DQA approaches\nimplemented in six US data-sharing networks. Other objectives of this review include identifying DQA programs\nthat attempt to automate EHR DQA and the DQ problems\n\n## 1 3\n\n\n-----\n\ninvestigated by these programs and developing a conceptual\nexplanation of the relationships between identified features\nand components.\n\n### Methods\n\n#### Search strategy and information sources\n\nIn this review, relevant articles published between February\n2011 and February 2021 that discuss EHR DQA were examined using the Preferred Reporting Items for Systematic reviews\nand Meta-Analyses extension for Scoping Reviews (PRISMAScR) guideline. The articles were identified through a comprehensive search of three electronic bibliographic databases:\nPubMed, Web of Science, and Scopus, using the queries below:\n\n1. (“information system” OR electronic OR computerised)\n2. (medical OR health OR clinic OR hospital OR patient)\n3. (“data quality” OR “data validation” OR “data integrity”\nOR “data error” OR “data completeness” OR “data consistency” OR “data accuracy” OR “data correctness” OR\n“data currency” OR “data plausibility”)\n4. 2011–2021 (February 2021).\n\nKeywords for the queries were drawn after a series of\npreliminary trial searches that considered the search strategies employed in related studies [10, 49, 51]. Reference lists\nof included papers were also checked using our eligibility\ncriteria for articles not captured in our initial search.\n\n#### Eligibility criteria and study selection\n\nArticles included in this review were selected based on the\nfollowing criteria: (1) describe a computerised DQA program not specific to the preference of an individual user\nor study, (2) target data from an EHR system, and (3) be\npublished in English. Articles that report assessments of\nhealth surveys, regional health statistics, clinical trials, and\nother health records not directly sourced from an EHR were\nexcluded. One reviewer [OO] screened the titles and abstracts\nof 1841 articles from the literature searches and the full text\nof 116 relevant titles and abstracts. Of these, 26 articles were\nselected for a full review. [OO] and [AH] each reviewed all\nthe 26 articles selected, while [PS] reviewed 25% (randomly\nselected). Disagreements were resolved by consensus, and\nthree (n = 3) studies were excluded because they provided\nlittle detail about their approach or context. Figure 3 presents\na flow diagram showing our search strategy and results.\n\n## 1 3\n\n\n#### Data extraction and analysis\n\nFor each article included in this review, relevant data\nwere abstracted using an Excel template. The data items\nabstracted include the author’s name, year of publication,\nand the name and description of the DQA program discussed. Other data items captured include the data error (DQ\ndimension) investigated, the context of the DQA implementation, the geographical location (country) and other designrelated features and considerations. Data errors investigated\nwere harmonised using Kahn et al. [11] definitions, cited\nnumerous times by related studies.\nLike previous related studies [10, 11, 52], we adopted an\ninductive and iterative approach in abstracting and codifying features and relevant considerations identified from the\narticles. An expanded literature review was also conducted\nto help refine specified features; in addition to the articles\nselected from the systematic search above, other articles\ndiscussing aspects relevant to developing or implementing\nDQA programs were reviewed, including materials such as\nDQ checks (rules) from large scale implementations [50,\n53], DQ frameworks and published best-practices [29–31,\n54–57], including those designed especially for EHR data\n\n[10, 11, 32, 43, 52, 58–63]. These additional materials were\nidentified using Google Scholar web searches and manual\nsearches of references in included studies.\n\n### Results\n\n#### Study summary and context\n\nWe identified 23 articles describing dedicated DQA programs implemented between 2013 and 2021, some of\nwhich have been deployed in real-world settings to assess\nEHR data quality (n = 14) [64–77], and a few experimental\nprototypes (n = 9) [48, 78–85], as shown in Table 1. Most\nof the DQA programs reported are affiliated with institutions in the USA (n = 12) and other countries, such as\nthe UK, Canada, Germany, Belgium, the Netherlands, and\nKenya. These DQA programs were designed for various\nuse cases, such as validating if data captured in particular\nEHRs conform to local system specifications [67, 72–74,\n76] or if they agree with data collected in other EHRs or\nother health information systems [64, 75]. Also, some of\nthe reported DQA programs focus on preparing datasets\nfor research studies [48, 66, 79, 83] and validating that\ndata from contributing sites conform to research network or\ndata warehouse specifications [65, 68–71, 77]. Only a few\n\n\n-----\n\n**Fig. 3 PRISMA-ScR flow diagram showing search strategy**\n\nappear generic and can be applied to different settings, data\ntypes and stages in the EHR data cycle [79, 82, 84, 85].\n\n#### Design features and considerations\n\nWe identified 24 features and considerations necessary for operationalising EHR DQA. These features have been grouped under\nfive top-level categories that include: defining DQA tasks (DQ_Task), acquiring and managing measurements (decision-making_\ncriteria) and other computational resources used to evaluate\ndefined DQA tasks (DQ-Measurement), collecting and managing target data (Target-Data), mechanisms for implementing\nmeasures (DQ-Mechanisms), and disseminating outcomes (DQ_Report) as shown in Table 2. We describe these categories and_\ntheir interrelationships in Fig. 4 and in the following subsections.\n\n\n**DQ‑task**\n\nThis category describes the specifications for the DQA activity, which may be defined formally or informally by Stake_holders, internal or external, to the individual or organisation_\nconducting the DQA activity, such as data consumers, program developers, data producers and host institutions [30].\nTypical items in a DQ-Task include specifications directly\nrelated to quality, such as the dimensions to investigate (DQ_Dimension), the data elements of interest, and a metric or_\nbaseline for ascertaining whether a dataset is good enough\nfor the intended use case (DQ-Metric). DQ-Task specification may also include non-functional specifications such as\nhow it should be conducted, such as the Periodicity at which\na DQA activity should be conducted, such as if it should be\n\n## 1 3\n\n\n-----\n\n**Table 1 Study summary**\n\n**Study** **Mechanism or Tool** **Description** **DQ dimension** **Year Location**\n**(Kahn et al. [11] equivalent)**\n\n\nÁlvarez Sánchez et al. [84] TAQIH A web-based data exploration completeness, value\ntool conformance, atemporal\nplausibility, uniqueness\n\n\n2019 Spain\n\n\nBotts et al. [64] NIST CDA validator Toolkit for verifying the\nconformance of exchanged\ndata to health information\nexchange standard\n\n\nvalue conformance 2014 US\n\n\nDaymont et al. [48] DQA toolkit R-based toolkit for assessing atemporal plausibility 2017 US\npaediatric growth data\n\n\nEstiri et al. [66] DQ[e] -c + Vue Toolkit for assessing\ncompleteness in a clinical\ndata research network\n\nEstiri et al. [80] DQ[e] -c Toolkit for assessing\ncompleteness in a clinical\ndata repository\n\nHart and Kuo [67] Island Health DQA Island Health Home and\nCommunity Care DQA\nImplementation\n\nHuser et al. [68] ACHILLES Heel An open-source software that\nprovides a useful starter set\nof rules for preparing data for\nuse in a CDRN\n\n\ncompleteness 2019 US\n\ncompleteness 2018 US\n\nDefined by the user 2017 Canada\n\nvalue conformance, 2016 US\nplausibility\n\n\nJohnson et al. [79] DQA toolkit Python implementation of the completeness, atemporal\nHDQF framework plausibility, value\nconformance\n\n\n2019 US\n\n\nJuárez et al. [69] QR generator A toolkit for validating data in\nlocal data warehouses in a\ndistributed research network\n\n\nDefined by the user 2019 Germany\n\n\nKapsner et al. [70] DQA Toolkit (R) A toolkit for preparing data for conformance, completeness 2019 Germany\nuse in a research network and plausibility\n\n\n2019 US\n\n\nKhare et al. [71] PEDSnet Data Quality Software implementation\nof the DQA program at\nPEDSnet CDRN\n\nLack et al. [72] DQA toolkit (C++) Software implementation of a\nDQA program for detecting\nerrors in treatment planning\nworkflows at a health facility\n\nMonda et al. [73] Extended OpenMRS DQA module implemented\nwithin an openMRS EHR\nsoftware\n\nNasir et al. [81] DCAP A tool for determining the\ncompleteness of individual\npatient records\n\n\ncompleteness, plausibility,\nvalue conformance, relational\nconformance\n\n\nconformance, completeness 2018 US\nand plausibility\n\nDefined by the user 2013 Kenya\n\ncompleteness 2016 US\n\n\nNoselli et al. [85] MonAT A web-based data exploration completeness, plausibility, 2017 UK\ntool value conformance\n\n\nQualls et al. [65] Self-contained package A package containing DQ\nanalysis programs for\nnetwork partners within the\nPCORnet DRN\n\nRabia et al. [74] DQA Toolkit Rule-based implementation of\na DQA program for assessing\ndischarge summaries\n\nRanade-Kharkar et al. [75] HIE Data Adjudicator Toolkit for assessing the\nquality of data entering or\nleaving a health information\nexchange framework\n\n## 1 3\n\n\nconformance, completeness 2018 US\nand plausibility\n\ncompleteness, atemporal 2018 Algeria\nplausibility\n\nplausibility, completeness 2014 US\n\n\n-----\n\n**Table 1 (continued)**\n\n**Study** **Mechanism or Tool** **Description** **DQ dimension** **Year Location**\n**(Kahn et al. [11] equivalent)**\n\n\nSilva et al. [82] DICOM Validator A web service for validating\nthe conformance of EHR\ndata produced by PACS to\nDICOM standards\n\nTute et al. [78] openCQA A DQA tool that uses\nopenEHR specifications\nto enable interoperable\nassessments\n\nvan der Bij et al. [76] DQ Feedback tool A feedback tool that evaluates\ndifferences in EHR data\namong practices and software\npackages\n\nVanbrabant et al. [83] DAQAPO-package A toolkit based on R\nthat enables automated\nassessment of EHR data\nfor emergency department\nsimulations\n\nWalker et al. [77] QA program (‘emrAdapter’) A toolkit for validating data in\nlocal data warehouses in the\nCER research network\n\n\nvalue conformance 2019 Portugal\n\nDefined by the user 2021 Germany\n\nconformance, completeness 2017 Netherlands\n\n\ncompleteness, temporal\nplausibility, atemporal\nplausibility, uniqueness\n\n\n2019 Belgium\n\n\nvalue conformance, 2014 US\nplausibility, completeness\n\n\nconducted on-demand [68, 83], autonomously or at set intervals, e.g., at the end of every day [65, 72].\nGenerally, a DQ-Task aims to assess one or more DQ_Dimensions in a given dataset, which could be a measurable_\nquality property, a collection of related measurements, or\ndatabase queries such as those used for many retrospective\nanalyses like case identification [10, 11, 42]. As observed\nin this study, the definitions of these DQ-Dimensions often\nvary and are derived from disparate sources, including\ndomain experts, literature reviews, and previous records\nof errors [65, 71, 72]. Some national bodies and research\ncommunities also prescribe _DQ-Dimension definitions_\nfor specific intended use cases, like intervention monitoring and aggregating data into federated networks [65, 70,\n76, 77, 86]. Also, given the increasingly task-dependent\nrequirement of many DQA activities, some programs allow\ndata consumers to specify the DQ-Dimensions they want\nto investigate dynamically at runtime [69, 73]. These definitions can be expressed in different formats, including\nnatural language (text), ontologies [52], taxonomies [39,\n74, 83], queries [77, 87], and other computational structures. Terms used to describe _DQ-Dimension may also_\nhave multiple connotations. For example, completeness is\na typical DQ-Dimension instance. The term has been used\nto describe records with missing values, values not in the\ndesired formats, or data elements with insufficient information (predictive value) for the intended use [80, 88]. DQA\nprograms with the additional requirement of comparing\n\n\noutcomes, root cause analysis, and implementing improvements might find this ambiguity property problematic.\nSelecting the DQ-Dimensions to assess is another critical consideration in defining DQ-Tasks as it indicates the\ncoverage of the DQA activity and the type of measurements\n(DQ-Measurement) that will be required. In some instances,\nthe DQ-Dimension selected may also determine targetable\ndata elements and levels in a DQA activity because certain\n_DQ-Measurements may only be applicable for data elements_\nof a particular domain, data type, and level [56, 65, 89].\nSimilarly, it is unlikely that DQA programs will be able to\nevaluate all possible DQ-Dimensions against all available\ndata elements, especially for large EHR datasets, which often\nhave diverse stakeholders. Some required DQ-Measurements\nmay be unavailable or too complicated to operationalise\n\n[49, 90]. Equally, datasets with many attributes, complex\ndata types, such as images, and large sizes, could demand\nmore resources beyond the mechanism (DQ-Mechanism)\navailable to execute the DQ-Task. So, for such scenarios,\ntrade-offs between DQ-Dimensions, data elements, time, and\ncapability of the DQ-Mechanism are essential to improving the efficiency of the DQA activity. Examples of such\ntrade-offs could include focusing on data elements necessary for intended use cases [72, 76, 91], those prevalent in\nthe previous records [61] and literature reviews [61, 84],\nor having more weight regarding their contributions to the\noverall quality of a dataset [92]. A DQ-Task may also be\nlimited to DQ-Dimensions that are feasible to investigate,\n\n## 1 3\n\n\n-----\n\n## 1 3\n\n\n-----\n\n## 1 3\n\n\n-----\n\n**Fig. 4 UML representation of concepts for operationalising EHR data quality assessments**\n\n\ni.e., required measurements and data are available [61, 65] or\ndata elements with a high return on investment (the tendency\nof finding data elements in most datasets) [88]. In this study,\nwe have used the term Priority to represent such trade-offs\nand their rationale. Capturing this information is essential\nfor transparency. It also helps to ensure that organisations’\nDQA coverage expands progressively.\nFurthermore, depending on the intended goal, a DQ-Task\nmay include a metric and a baseline for determining if the\ntarget dataset is good enough for the intended use case. This\nmetric (DQ-Metric), which is an aggregate score, could be\nquantitative (e,g., count [66, 77], simple ratio [73, 80], percentage), categorical (e.g., ordinal, Boolean [73]) or other complex\nmetrics [54, 93]. As inferred from this review, these metrics are\napplied to aggregated outcomes of DQ-Measurements across\ndifferent data levels (field, record, table). They help present\nassessment results in easily digestible and comparable formats\n\n[13, 68] and may be embedded as part of DQ-Measurements\ngiven their close associations.\n\n**DQ‑measurement**\n\nThis category refers to the criteria for evaluating selected\n_DQ-Dimensions. It typically encompasses one or more com-_\nparisons involving data elements’ content, derivation, property (e.g., type, format) and reference items across different\ndata levels (cell, record, table). In this review, target data elements are subsets of the data elements defined in the DQ-Task\n\n## 1 3\n\n\ndefinition and the data model. A data model is described\nin the next section. Reference items can be any values held\nin other data elements in the same dataset, the outcome of\nother DQ-Measurements and explicitly defined values, like\nnumbers, Boolean, text, value ranges, regular expression, and\nvalue sets [10]. The data type of the data element evaluated\nmay determine the kind of reference item required. For example, range and spelling checks would likely be used to assess\ndata elements of type numeric and text.\nCommon comparisons include assessing value conformance, such as values presence, conformance to defined patterns, precision, allowable ranges or value sets, functional\ndependencies and causal relationships [10, 11]. It may also\ninvolve evaluating agreement with other data sources like\na previous snapshot of the same data, other datasets within\nthe same or different EHR systems, and recollected observations [94, 95]. For DQ-Measurements involving disparate\ndatasets, it is essential to note that the datasets may have\nsyntactical and semantical differences. And while various\ntransformation functions and tools exist to normalise datasets, excessive transformations can overestimate or underestimate DQ-Dimensions.\nFurthermore, as stated earlier, DQ-Measurements apply\nto specific data levels (cell, field, record, table) [11, 43, 50,\n56, 89]. For instance, in assessing value conformance, DQ_Measurements may target single data cells in records, such_\nas checking if single data cells match specifications like\ndata type and format [67, 88, 89]. In the same way, some\n\n\n-----\n\n_DQ-Measurements apply to the field level, comparing the_\noutput of aggregating selected observations (records) in that\nfield with reference information, such as identifying univariate outliers and evaluating redundancy [43, 68]. Others\ninvolve multiple data elements across a record level, such as\nidentifying functional dependency violations [39, 83] and\nagreement between multiple variables like fields containing\ndiagnoses and medication concepts [50, 53]. Likewise, multiple data elements can also be compared across aggregated\nrecords, such as comparing the value of a data element with\nsuccessive values of the same fields for a given subject to\ndetermine if values changed implausibly over time. It is also\npossible for DQ-Measurements to act on the table level and\nfor multiple DQ-Measurements to be combined using logical\njunctions like AND, OR, and NOT to investigate complex\n_DQ-Dimensions [67]._\nLike DQ-Dimensions, the logic for DQ-Measurements\nmay be acquired from multiple knowledge sources, including domain experts [68–72, 79, 80], data consumers [69, 73],\ncrowdsourcing [53, 68, 71], data standards or dictionaries\n\n[71, 77, 78], national guidelines [44], literature review [71],\nand other existing DQ-Measurements [53]. Studies have\nalso shown that it is possible to create DQ-Measurements\ninductively from datasets using statistical measures, natural language processing (NLP), machine learning and rule\nmining techniques, which also offer automated capabilities\n\n[96–100]. Nevertheless, acquiring DQ-Measurements from\nthese sources may involve varying confidence, coverage, and\nacquisitional efficiency. For instance, domain experts can\nproduce DQ-Measurements via interviews and crowdsourcing, which may command high confidence locally, but could\nalso be expensive, time-consuming and have low coverage\n\n[97, 101]. Likewise, DQ-Measurements developed using\ndata-driven techniques can be inconsistent, unexplainable,\nand prone to false positives.\n\n**Target‑data**\n\nThis category encompasses considerations in handling input\ndata in a DQA activity, including how it will be accessed,\nsupported formats, and data storage. Some methods reportedly used for accessing EHR data for assessment include\ndirect execution of database scripts and accessing health\ninformation exchange frameworks like openEHR [78].\n_Target-Data have also been extracted from EHR reposito-_\nries and made available in filesystem formats like commaseparated-values (CSV) [84, 87]. The approach employed to\naccess EHR data is often determined by host environments,\ndata protection policies, infrastructure, performance, and\ninterconnectivity. For instance, some institutions require\nEHR data to be accessed remotely to enable more autonomy\nover their data and reduce the likelihood of security and\nprivacy breaches [77, 78]. Size is another factor, as it is not\n\n\nalways timely, economical, or safe to inspect every record\nin a given data source [58, 63, 68, 79]. So, instead of assessing the whole dataset, subsets of the original data may be\nselected using sampling and randomisation strategies [40,\n58]. However, assessment outcomes do not always reflect\nthe dataset’s DQ status. Also, determining the appropriate\ndataset size sufficient to estimate the state of the whole dataset can be challenging [58].\nFurthermore, a DQ-Task could also entail comparing\nor assessing Target-Data that use different syntactical and\nsemantical standards to store data. To help ensure all components operate and share data uniformly, some DQA programs employ Common Data Models (CDMs). Examples\nof commonly reported CDMs include the Observational\nMedical Outcomes Partnership (OMOP) CDM [68, 71,\n80], Sentinel CDM (SCDM) [50], Informatics for Integrating Biology & the Bedside (i2b2) [70], Digital Imaging\nand Communications in Medicine (DICOM) [72, 82], and\nopenEHR [78]. These CDMS contain varying data elements\ndefined for a particular aggregated form, institution, or use\ncase [68, 70, 78, 80] and linked differently [60, 77, 102]. In\nmost instances, only a single CDM is supported, which is,\napparently, more straightforward to implement. However,\nthis approach limits DQA programs and makes them not\ngeneralisable and scalable to other sites [80, 103]. With\nmore institutions exchanging and aggregating data, there\nwould likely be more demand for DQA programs to support multiple data structures and study designs.\nSimilarly, EHR data are not always in the same structure\nas the specified CDM. In such scenarios, data integration is\nrequired. Common approaches for integrating data sources\ninclude extraction transformation and loading (ETL) activities, data replication, or a virtual representation [104]. These\ndata integration activities often require pre-defined schema\nmappings of source and target data models, which can be\nhardcoded, or defined dynamically using interactive interfaces, configuration files and other semi- or fully automated\napproaches [60, 104]. In addition, data transformation may\nbe required to convert source data, especially unstructured\ndata, to a format appropriate for target DQ-Measurements\n\n[29, 105].\n\n**DQ‑report**\n\nThis component refers to the content and verbosity of the\noutcomes from executing a DQ-Task. It provides feedback\nto enable stakeholders to judge their datasets, including\nremediation recommendations, which can trigger and shape\nimprovement efforts. For instance, a typical DQ-Report\ncontent may contain a collection of returned records that\nsatisfy the DQ-Dimensions evaluated, DQ-Metric scores and\nmetadata containing details of other concepts involved in\nthe DQA process, including possible enhancements. These\n\n## 1 3\n\n\n-----\n\noutcomes can be communicated to Stakeholders using a\npreferred Dissemination-Method like tables and graphs that\nallow for quick analysis and provide visual attributes for\ndrawing attention to specific results and details. DQ-Report\ncan also be exported to relevant bodies or stored for further analysis. Similarly, Dissemination-Methods may also\nincorporate features that enable them to fulfil reporting\nrequirements, such as interface designs, password protection, anonymisation functions and secured data transfers, as\ndiscussed below.\n\n**DQ‑mechanism**\n\nThis category refers to the program, process or tools\nemployed to operationalise the different activities involved\nin executing a DQ-Task and the features that enable them to\nperform satisfactorily and meet stakeholders’ requirements.\nCommonly reported features identified in this review have\nbeen grouped under configurability, usability, scalability,\nperformance, and security, as shown in Table 3 below.\n\n### Discussion\n\nThis review examines recent efforts to automate EHR DQA.\nSo far, we have identified 23 DQA programs, with more than\n80% implemented within the last five years (at the time of\nthe search). This trend shows organisations using EHR data\nfor analysis are becoming more aware of the inherent quality\nproblems. It also affirms the growing focus on automating\nEHR DQA, driven mainly by the need to help researchers\nprepare EHR data to meet research objectives. However,\n\n**Table 3 Example of DQ-Mechanism features**\n\n\nonly a few DQA programs currently focus on improving the\ndata quality at source EHRs, which is critical for preventing\nimmediate medical and operational mishaps and improving\nelectronic documentation.\nThe latter can be attributed to available DQA programs\nnot being as robust as desired, focusing on DQ dimensions, such as completeness and value conformance, which\ncan be considered trivial to implement and are currently\nbeing supported by various data integration and analytic\ntools. Also, unstructured data formats like free text and\nimages, which make up most data stored in EHRs [7], are\ncomputationally more challenging to analyse [8] and hence\nrarely supported. Similarly, many of the reported DQA\nprograms are tightly coupled to existing infrastructure\nand are available only to users of the same community.\nSome of them are also too technical, lack interactivity and\nrequire users to know about the host systems and supported programming languages, like knowledge of R, to\noperate the DQA tool [66, 68]. They are also not being\nevaluated adequately; hence, they are not ready for general\nclinical use.\nThese limitations further emphasise the challenge of\nconducting EHR DQA. Interestingly, our extended review\nshowed no lack of frameworks discussing DQ theories,\nbest practices, and other concepts associated with DQA.\nFor instance, several frameworks like the Total Data Quality Management (TDQM) framework describe best practices for improving overall DQ and conducting DQA from\na general perspective [29, 30, 54–57] and a few others tailored explicitly for EHR data [10, 11, 32, 52]. However, it\nis unclear how the many theoretical concepts can be translated into practice, amongst other factors. For example,\n\n\n**Feature** **Description** **Examples**\n\n**Configurability** Allow users to personalise, adapt or extend the DQA process to match their _DQ-Dimension [73, 78]_\nrequirements or environment Weights in DQ-Metric [81, 84]\n_Baselines [84]_\n_DQ-Measurement [69, 73]_\n_Schema-Mapping [69, 80, 81]_\n**Usability** Enable users to perform tasks efficiently and effectively Graphic user interface [66, 84]\nInteractive options [66]\n**Scalability** Enable the system to sustainably respond to changes in resource demand, datasets, or Modular design [80]\nenvironment Multi Data-Model support [80]\nInteroperable DQ-Measurement [78]\nFast deployment [70]\n**Performance** Enable the system to maintain satisfactory levels of responsiveness and stability for Parallel computing\nspecified workloads Out-of-memory execution [80]\nBatch processing [77, 82]\nSampling strategy [58, 77]\n**Security** Ensure security concerns such as privacy and proprietary protections are satisfied Password protection [66, 80]\nSecured file sharing [66]\nDe-identification [72, 76, 82]\n\n## 1 3\n\n\n-----\n\nmany existing frameworks focus on standardising DQ\ndimensions and identifying potential assessment methods,\nbut they do not provide much regarding how these methods\ncan be operationalised in real-world settings. Also, only\na handful of studies investigate other critical aspects of\nDQA, such as data management [43, 58, 60] and reporting and applying outcomes [59, 63]. The concepts are also\ndiscussed in isolation and, thus, contain competing and\nambiguous terms, which introduce confusion and make it\ndifficult to translate them into practice [38, 80].\n\n#### Strengths and limitations\n\nThis study identifies several programs and tools developed,\nimplemented, or adopted for automating EHR DQA using\na systematic approach. In addition to previous studies using\nthis approach, our choice was also motivated by the benefits\nof not limiting our analysis to the authors’ preconceptions\nand the ability to organise information and assumptions\nexplicitly. However, the list of DQA programs identified may\nnot be exhaustive as we focused on only those published in\nselected bibliographic databases. Unpublished programs or\nthose available to select users, including proprietary programs, were outside the scope of this review.\nNonetheless, this review identified several critical components and considerations in developing and operationalising\nDQA programs for EHR data. These components have been\ngrouped under five top-level categories: defining DQ tasks,\ndeveloping and managing measurements for inspecting\ndatasets, collecting and handling target datasets for assessment, analysing and disseminating outcomes, and mechanisms for operationalising all these components. As shown\nin Fig. 4, we have explained these categories extensively\nusing UML diagram concepts and domain-independent\nterms derived from standard ontologies, like the Basic Formal Ontology [106] and other reviewed frameworks, in our\nattempt to disambiguate the so-called complex activity of\nconducting EHR DQA. The components identified have also\nbeen organised to reflect expected knowledge requirements\nand practicality. This is intended to foster better collaboration between stakeholders, such as data owners, reporting\nteams, and knowledge curators, and encourage the reuse\nof resources like data integration tools, rule engines, and\nreporting frameworks. It also allows each component to be\nstandardised individually against having one general standard. Furthermore, we anticipate that the identified concepts\ncan help to curate knowledge of the different approaches to\nDQA, which is a bold step toward standardising health data\nquality assessment, as demonstrated in Table 2.\nThis work has some similarities with existing works and\nsome essential additions, even though expressed in different\nlanguages in some cases. For instance, it recognises the taskdependent nature of DQA and the importance of a well-defined\n\n\nplan [50, 52]. In addition to specifying DQ dimensions to\nassess, it notes that how assessments are conducted shapes\nthe scope and contributes to the variability of DQA processes,\nsuch as the periodicity of checks and prioritisation strategy.\nSimilarly, while there is no unique way of measuring DQ\ndimensions, this review explicitly expounds on the structure\nand complexities involved in developing and managing DQ\nmeasurements, which could help reduce the confusion surrounding the development of new assessment methods. In\naddition, this works attempts to propose a relationship between\nDQ concepts and attributes, which have been mentioned in\nisolation in various existing works, as shown in Fig. 4.\nNonetheless, this review has a task-centric focus, emphasising technological-related components reported in the literature. Also, while we took great care to ensure that the\nliterature search was broad and systematic, our findings may\nbe missing some necessary components not discussed in the\narticles reviewed. This study did not also elicit the views of\nthe different EHR data users to validate the findings from\nthis review. So, while our results reflect shared conceptualisations across the literature and considerations that could\napply uniformly, further research may benefit from more\nvalidation, including obtaining stakeholder input on the utility of our contribution in practice.\n\n### Conclusion\n\nEHR data are a critical component of today’s healthcare\nindustry and must be good enough to support clinical care\nor other secondary use cases. Various strategies have been\nproposed to ensure this, including DQA activities for detecting problems that need attention. Nevertheless, anecdotal\nevidence suggests an absence of comprehensive tools for\nfacilitating reliable and consistent assessments. In light of\nthis, we have examined the literature in this study to assess\nthis gap and identify important considerations for developing and implementing new DQA tools. Our findings show\nthat automating EHR DQA is gaining traction. However,\nthere appears to be a general lack of clarity surrounding\nDQA processes brought about by the contextual nature of\nDQ requirements, heterogeneity of EHR data, and the challenge of developing measurements for inspecting datasets.\nMore worrisome is that the quality of these processes is\nunknown as, in many cases, they are not backed by theoretical frameworks, and there are no obligations to certify\nthat DQA tools measure what they are designed to measure.\nThere is also a growing demand for interoperable checks that\napply to multiple contexts. Healthcare organisations hoping\nto develop DQA programs will find this review helpful as we\nhave summarised what exists and shed light on critical components required to operationalise DQA processes. We also\n\n## 1 3\n\n\n-----\n\nanticipate that this work would help reduce the confusion\naround EHR data quality management and provide guidance\nappropriate for developing effective programs.\n\n**Funding The University of Portsmouth Global PhD Full Scholarship**\nsponsored this study.\n\n**Data availability (data transparency) The datasets generated and ana-**\nlysed during the current study are available from the corresponding\nauthor on reasonable request.\n\n**Code availability (software application or custom code) Not applicable.**\n\n#### Declarations\n\n**Conflict of interest The authors also have no conflicts of interest to**\ndeclare relevant to this article’s content.\n\n**Open Access This article is licensed under a Creative Commons Attri-**\nbution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long\nas you give appropriate credit to the original author(s) and the source,\nprovide a link to the Creative Commons licence, and indicate if changes\nwere made. The images or other third party material in this article are\nincluded in the article's Creative Commons licence, unless indicated\notherwise in a credit line to the material. If material is not included in\nthe article's Creative Commons licence and your intended use is not\npermitted by statutory regulation or exceeds the permitted use, you will\nneed to obtain permission directly from the copyright holder. To view a\n[copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/.](http://creativecommons.org/licenses/by/4.0/)\n\n### References\n\n1. Warren, L.R., et al., Improving data sharing between acute hospi_tals in England: an overview of health record system distribution_\n_and retrospective observational analysis of inter-hospital transi-_\n_tions of care. BMJ Open, 2019. 9(12): p. e031637._\n\n2. Atasoy, H., B.N. Greenwood, and J.S. McCullough, The Digi_tization of Patient Care: A Review of the Effects of Electronic_\n_Health Records on Health Care Quality and Utilization. Annu_\nRev Public Health, 2019. 40: p. 487–500.\n\n3. Hayrinen, K., K. Saranto, and P. Nykanen, Definition, struc_ture, content, use and impacts of electronic health records: a_\n_review of the research literature. Int J Med Inform, 2008. 77(5):_\np. 291–304.\n\n4. Meystre, S.M., et al., Clinical Data Reuse or Secondary Use:\n_Current Status and Potential Future Progress. Yearb Med_\nInform, 2017. 26(1): p. 38–52.\n\n5. Goldstein, B.A., et al., Opportunities and challenges in develop_ing risk prediction models with electronic health records data:_\n_a systematic review. J Am Med Inform Assoc, 2017._ **24(1):**\np. 198–208.\n\n6. Topol, E., The Topol Review Preparing the Healthcare Workforce\nto Deliver the Digital Future, 2019: p. 1–48.\n\n7. Safran, C., Update on Data Reuse in Health Care. Yearb Med\nInform, 2017. 26(1): p. 24–27.\n\n8. Jensen, P.B., L.J. Jensen, and S. Brunak, Mining electronic health\n_records: towards better research applications and clinical care._\nNat Rev Genet, 2012. 13(6): p. 395–405.\n\n9. Vuokko, R., et al., Secondary Use of Structured Patient Data:\n_Interim Results of A Systematic Review. 2015. p. 291–295._\n\n## 1 3\n\n\n10. Weiskopf, N.G. and C. Weng, Methods and dimensions of elec_tronic health record data quality assessment: Enabling reuse for_\n_clinical research. Journal of the American Medical Informatics_\nAssociation, 2013. 20(1): p. 144–151.\n\n11. Kahn, M.G., et al., A Harmonized Data Quality Assessment Ter_minology and Framework for the Secondary Use of Electronic_\n_Health Record Data. EGEMS (Wash DC), 2016. 4(1): p. 1244._\n\n12. Bayley, K.B., et al., Challenges in using electronic health record\n_data for CER: experience of 4 learning organizations and solu-_\n_tions applied. Med Care, 2013. 51(8 Suppl 3): p. S80-6._\n\n13. WHO, Administrative Errors: Technical Series on Safer Primary\n_Care, in WHO Press. 2016: Geneva._\n\n14. Ghosh, A., S. McCarthy, and E. Halcomb, Perceptions of primary\n_care staff on a regional data quality intervention in Australian_\n_general practice: A qualitative study. BMC Family Practice,_\n2016. 17(1).\n\n15. Collins, S.A., et al., Clinician preferences for verbal commu_nication compared to EHR documentation in the ICU. Applied_\nClinical Informatics, 2011. 2(2): p. 190–201.\n\n16. Salomon, R.M., et al., Openness of patients’ reporting with use of\n_electronic records: Psychiatric clinicians’ views. Journal of the_\nAmerican Medical Informatics Association, 2010. 17(1): p. 54–60.\n\n17. Peivandi, S., et al., Evaluation and comparison of errors on nurs_ing notes created by online and offline speech recognition tech-_\n_nology and handwritten: an interventional study. BMC Medical_\nInformatics and Decision Making, 2022. 22(1): p. 96.\n\n18. Colin, N.V., et al., Understanding the Impact of Variations in\n_Measurement Period Reporting for Electronic Clinical Quality_\n_Measures. EGEMS (Wash DC), 2018. 6(1): p. 17._\n\n19. Bowman, S., Impact of electronic health record systems on infor_mation integrity: quality and safety implications. Perspect Health_\nInf Manag, 2013. 10: p. 1c-1c.\n\n20. O’Donnell, H.C., et al., Physicians’ Attitudes Towards Copy and\n_Pasting in Electronic Note Writing. Journal of General Internal_\nMedicine, 2009. 24(1): p. 63–68.\n\n21. Coleman, N., et al., From patient care to research: A valida_tion study examining the factors contributing to data quality in a_\n_primary care electronic medical record database. BMC Family_\nPractice, 2015. 16(1).\n\n22. _Economic analysis of the prevalence and clinical and economic_\n_burden of medication error in England BMJ Quality & Safety,_\n2021. 30(2): p. 96.\n\n23. Zozus, M.N., M. Penning, and W.E. Hammond, Factors impact_ing physician use of information charted by others. JAMIA_\nOpen, 2019. 2(1): p. 107–114.\n\n24. Munyisia, E.N., D. Reid, and P. Yu, Accuracy of outpatient ser_vice data for activity-based funding in New South Wales, Aus-_\n_tralia. Health Inf Manag, 2017. 46(2): p. 78–86._\n\n25. Kaplan, B., How Should Health Data Be Used?: Privacy, Sec_ondary Use, and Big Data Sales. Cambridge Quarterly of Health-_\ncare Ethics, 2016. 25(2): p. 312–329.\n\n26. Nouraei, S.A.R., et al., Accuracy of clinician-clinical coder infor_mation handover following acute medical admissions: Implica-_\n_tion for using administrative datasets in clinical outcomes man-_\n_agement. Journal of Public Health (United Kingdom), 2016._\n**38(2): p. 352–362.**\n\n27. Feldman, K., et al., Beyond volume: The impact of complex\n_healthcare data on the machine learning pipeline Lecture Notes_\nin Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2017.\n**10344 LNAI: p. 150–169.**\n\n28. Hanauer, D.A., et al., Complexities, variations, and errors of\n_numbering within clinical notes: The potential impact on infor-_\n_mation extraction and cohort-identification. BMC Medical Infor-_\nmatics and Decision Making, 2019. 19(Suppl 3): p. 75–75.\n\n\n-----\n\n29. Batini, C., et al., Methodologies for data quality assessment and\n_improvement. ACM computing surveys (CSUR), 2009. 41(3): p. 16._\n\n30. Wang, R.Y., A product perspective on total data quality manage_ment. Communications of the ACM, 1998. 41(2): p. 58–66._\n\n31. Veiga, A.K., et al., A conceptual framework for quality assess_ment and management of biodiversity data. PLoS ONE, 2017._\n**12(6): p. e0178731-e0178731.**\n\n32. Weiskopf, N.G., et al., A Data Quality Assessment Guideline\n_for Electronic Health Record Data Reuse. EGEMS (Wash DC),_\n2017. 5(1): p. 14.\n\n33. Kelly, M.M., R.J. Coller, and P.L. Hoonakker, Inpatient Portals\n_for Hospitalized Patients and Caregivers: A Systematic Review._\nJ Hosp Med, 2018. 13(6): p. 405–412.\n\n34. Wiebe, N., et al., Evaluation of interventions to improve inpa_tient hospital documentation within electronic health records:_\n_a systematic review. J Am Med Inform Assoc, 2019. 26(11):_\np. 1389–1400.\n\n35. Isaksen, H., et al., Design of tooltips for data fields: A field\n_experiment of logging use of tooltips and data correctness. 2017._\np. 63–78.\n\n36. Avidan, A. and C. Weissman, Record completeness and data\n_concordance in an anesthesia information management system_\n_using context-sensitive mandatory data-entry fields. Interna-_\ntional Journal of Medical Informatics, 2012. 81(3): p. 173–181.\n\n37. McCormack, J.L. and J.S. Ash, Clinician perspectives on the\n_quality of patient data used for clinical decision support: a_\n_qualitative study AMIA … Annual Symposium proceedings /_\nAMIA Symposium. AMIA Symposium, 2012. 2012(Table 1):\np. 1302–1309.\n\n38. Roomaney, R.A., et al., Availability and quality of routine\n_morbidity data: Review of studies in South Africa. Journal of_\nthe American Medical Informatics Association, 2017. 24(e1):\np. e194-e206.\n\n39. Zhang, Y. and G. Koru, Understanding and detecting defects in\n_healthcare administration data: Toward higher data quality to_\n_better support healthcare operations and decisions. Journal of_\nthe American Medical Informatics Association, 2020. 27(3):\np. 386–395.\n\n40. WHO, Data Quality Assessment of National and Partner Hiv\n_Treatment and Patient Monitoring Systems 2018(August):_\np. 1–68.\n\n41. Feder, S.L., _Data Quality in Electronic Health Records_\n_Research: Quality Domains and Assessment Methods. West J_\nNurs Res, 2018. 40(5): p. 753–766.\n\n42. Reimer, A.P., A. Milinovich, and E.A. Madigan, Data quality\n_assessment framework to assess electronic medical record data_\n_for use in research. International Journal of Medical Informat-_\nics, 2016. 90: p. 40–47.\n\n43. Kahn, M.G., et al., A pragmatic framework for single-site and\n_multisite data quality assessment in electronic health record-_\n_based clinical research. Med Care, 2012. 50 Suppl: p. S21-9._\n\n44. Muthee, V., et al., The impact of routine data quality assess_ments on electronic medical record data quality in Kenya._\nPLoS ONE, 2018. 13(4).\n\n45. Yadav, S., et al., Comparison of accuracy of physical exami_nation findings in initial progress notes between paper charts_\n_and a newly implemented electronic health record. Journal of_\nthe American Medical Informatics Association, 2017. 24(1):\np. 140–144.\n\n46. Abiy, R., et al., A Comparison of Electronic Medical Record\n_Data to Paper Records in Antiretroviral Therapy Clinic in_\n_Ethiopia: What is affecting the Quality of the Data? Online J_\nPublic Health Inform, 2018. 10(2): p. e212-e212.\n\n47. Maletic, J.I. and A. Marcus, Data Cleansing: Beyond Integrity\n_Analysis Iq, 2000: p. 1–10._\n\n\n48. Daymont, C., et al., Automated identification of implausible\n_values in growth data from pediatric electronic health records._\nJ Am Med Inform Assoc, 2017. 24(6): p. 1080–1087.\n\n49. Bian, J., et al., Assessing the practice of data quality evaluation\n_in a national clinical data research network through a system-_\n_atic scoping review in the era of real-world data. Journal of_\nthe American Medical Informatics Association, 2020. 27(12):\np. 1999–2010.\n\n50. Callahan, T.J., et al., A Comparison of Data Quality Assess_ment Checks in Six Data Sharing Networks. EGEMS (Wash_\nDC), 2017. 5(1): p. 8.\n\n51. Chen, H., et al., A review of data quality assessment methods\n_for public health information systems. Int J Environ Res Public_\nHealth, 2014. 11(5): p. 5170–207.\n\n52. _A Data Quality Ontology for the Secondary Use of EHR Data_\nAMIA … Annual Symposium proceedings. AMIA Symposium, 2015. 2015: p. 1937–1946.\n\n53. Wang, Z., et al., Rule-Based Data Quality Assessment and\n_Monitoring System in Healthcare Facilities. Stud Health Tech-_\nnol Inform, 2019. 257: p. 460–467.\n\n54. Pipino, L.L., Y.W. Lee, and R.Y. Wang, Data Quality Assessment\nCommunications of the ACM, 2002.\n\n55. Naumann, F. and C. Rolker, Assessment Methods for Information\n_Quality Criteria Information Systems, 2000: p. 148–162._\n\n56. Woodall, P., M. Oberhofer, and A. Borek, A classification of\n_data quality assessment and improvement methods. International_\nJournal of Information Quality, 2014. 3(4): p. 298–321.\n\n57. DAMA UK Working Group, The six primary dimensions for data\n_quality assessment: defining data quality dimensions 2013._\n\n58. Pageler, N.M., et al., A rational approach to legacy data valida_tion when transitioning between electronic health record systems._\nJournal of the American Medical Informatics Association, 2016.\n**23(5): p. 991–994.**\n\n59. Kahn, M.G., et al., Transparent reporting of data quality in dis_tributed data networks. EGEMS (Wash DC), 2015. 3(1): p. 1052._\n\n60. Hartzema, A.G., et al., Managing data quality for a drug safety\n_surveillance system. Drug Safety, 2013. 36(SUPPL.1): p. S49-S58._\n\n61. Terry, A.L., et al., A basic model for assessing primary health\n_care electronic medical record data quality. BMC Medical Infor-_\nmatics and Decision Making, 2019. 19(1).\n\n62. Rogers, J.R., et al., A Data Element-Function Conceptual Model\n_for Data Quality Checks. EGEMS (Wash DC), 2019. 7(1): p. 17._\n\n63. Callahan, T., et al., Reporting Data Quality Assessment Results:\n_Identifying Individual and Organizational Barriers and Solu-_\n_tions. EGEMS (Wash DC), 2017. 5(1): p. 16._\n\n64. _Data Quality and Interoperability Challenges for eHealth_\n_Exchange Participants: Observations from the Department of_\n_Veterans Affairs’ Virtual Lifetime Electronic Record Health Pilot_\n_Phase AMIA … Annual Symposium proceedings / AMIA Sym-_\nposium. AMIA Symposium, 2014. 2014: p. 307–314.\n\n65. Qualls, L.G., et al., Evaluating Foundational Data Quality in the\n_National Patient-Centered Clinical Research Network (PCOR-_\n_net®). EGEMS (Wash DC), 2018. 6(1): p. 3._\n\n66. Estiri, H., et al., A federated EHR network data completeness\n_tracking system. Journal of the American Medical Informatics_\nAssociation, 2019. 26(7): p. 637–645.\n\n67. Hart, R. and M.H. Kuo, Better Data Quality for Better Health_care Research Results - A Case Study. Stud Health Technol_\nInform, 2017. 234: p. 161–166.\n\n68. Huser, V., et al., Multisite Evaluation of a Data Quality Tool\n_for Patient-Level Clinical Data Sets. EGEMS (Wash DC), 2016._\n**4(1): p. 1239.**\n\n69. Juárez, D., et al., A Generic Method and Implementation to Evalu_ate and Improve Data Quality in Distributed Research Networks._\nMethods of Information in Medicine, 2019. 58(2–3): p. 86–93.\n\n## 1 3\n\n\n-----\n\n70. Kapsner, L.A., et al., Moving Towards an EHR Data Quality\n_Framework: The MIRACUM Approach. Stud Health Technol_\nInform, 2019. 267: p. 247–253.\n\n71. Khare, R., et al., _Design and Refinement of a Data Quality_\n_Assessment Workflow for a Large Pediatric Research Network._\nEGEMS (Wash DC), 2019. 7(1): p. 36.\n\n72. Lack, D., et al., Early detection of potential errors during patient\n_treatment planning. Journal of Applied Clinical Medical Physics,_\n2018. 19(5): p. 724–732.\n\n73. Monda, J., J. Keipeer, and M.C. Were, Data integrity module for\n_data quality assurance within an e-health system in sub-Saharan_\n_Africa. Telemed J E Health, 2012. 18(1): p. 5–10._\n\n74. Rabia, L., I.A. Amarouche, and K. Beghdad Bey. Rule-based\n_approach for detecting dirty data in discharge summaries. 2018._\n\n75. _Improving Clinical Data Integrity by using Data Adjudica-_\n_tion Techniques for Data Received through a Health Informa-_\n_tion Exchange (HIE) AMIA … Annual Symposium proceed-_\nings / AMIA Symposium. AMIA Symposium, 2014. **2014:**\np. 1894–1901.\n\n76. van der Bij, S., et al., Improving the quality of EHR recording in\n_primary care: A data quality feedback tool. Journal of the Ameri-_\ncan Medical Informatics Association, 2017. 24(1): p. 81–87.\n\n77. Walker, K.L., et al., Using the CER Hub to ensure data qual_ity in a multi-institution smoking cessation study. Journal of_\nthe American Medical Informatics Association, 2014. 21(6):\np. 1129–1135.\n\n78. Tute, E., I. Scheffner, and M. Marschollek, A method for interop_erable knowledge-based data quality assessment. BMC Medical_\nInformatics and Decision Making, 2021. 21(1).\n\n79. Johnson, S.G., et al., A Framework for Visualizing Data Quality\n_for Predictive Models and Clinical Quality Measures AMIA Jt_\nSummits Transl Sci Proc, 2019. 2019: p. 630–638.\n\n80. Estiri, H., et al., Exploring completeness in clinical data research\n_networks with DQe-c. Journal of the American Medical Informat-_\nics Association, 2018. 25(1): p. 17–24.\n\n81. Nasir, A., V. Gurupur, and X. Liu, A new paradigm to analyze\n_data completeness of patient data. Applied Clinical Informatics,_\n2016. 7(3): p. 745–764.\n\n82. Silva, J.M., et al., A community-driven validation service for\n_standard medical imaging objects Computer Standards and_\nInterfaces, 2019. 61(May 2018): p. 121–128.\n\n83. Vanbrabant, L., et al., _Quality of input data in emergency_\n_department simulations: Framework and assessment techniques_\nSimulation Modelling Practice and Theory, 2019. 91(December\n2018): p. 83–101.\n\n84. Álvarez Sánchez, R., et al., TAQIH, a tool for tabular data qual_ity assessment and improvement in the context of health data._\nComputer Methods and Programs in Biomedicine, 2019. 181:\n104824.\n\n85. Noselli, M., et al. MonAT: A visual web-based tool to profile\n_health data quality. 2017._\n\n86. Yoon, D., et al., Conversion and data quality assessment of elec_tronic health record data at a korean tertiary teaching hospital to_\n_a common data model for distributed network research. Health-_\ncare Informatics Research, 2016. 22(1): p. 54–58.\n\n87. Dziadkowiec, O., et al., Using a Data Quality Framework to\n_Clean Data Extracted from the Electronic Health Record: A Case_\n_Study. EGEMS (Wash DC), 2016. 4(1): p. 1201._\n\n## 1 3\n\n\n88. Weiskopf, N.G., et al., Defining and measuring completeness of\n_electronic health records for secondary use. Journal of Biomedi-_\ncal Informatics, 2013. 46(5): p. 830–836.\n\n89. _A Framework for Data Quality Assessment in Clinical Research_\n_Datasets AMIA … Annual Symposium proceedings. AMIA_\nSymposium, 2017. 2017: p. 1080–1089.\n\n90. Ehrlinger, L., E. Rusz, and W. Wöß, A survey of data qual_ity measurement and monitoring tools arXiv preprint arXiv:_\n1907.08138, 2019.\n\n91. Daniel, C., et al., Initializing a hospital-wide data quality pro_gram. The AP-HP experience Computer Methods and Programs_\nin Biomedicine, 2019. 181: 104804.\n\n92. Welch, G., et al., Data Cleaning in the Evaluation of a Multi-Site\n_Intervention Project. EGEMS (Wash DC), 2017. 5(3): p. 4–4._\n\n93. Huser, V., et al., Extending Achilles Heel Data Quality Tool with\n_New Rules Informed by Multi-Site Data Quality Comparison._\nStud Health Technol Inform, 2019. 264: p. 1488–1489.\n\n94. Liao, T.V., et al., Evaluation of medication errors with implementation\n_of electronic health record technology in the medical intensive care_\n_unit. Open Access Journal of Clinical Trials, 2017. 9: p. 31–40._\n\n95. _Concordance of Electronic Health Record (EHR) Data Describ-_\n_ing Delirium at a VA Hospital AMIA … Annual Symposium_\nproceedings / AMIA Symposium. AMIA Symposium, 2014.\n**2014: p. 1066–1071.**\n\n96. Estiri, H. and S.N. Murphy, Semi-supervised encoding for outlier\n_detection in clinical observation data. Comput Methods Pro-_\ngrams Biomed, 2019.\n\n97. Estiri, H., J.G. Klann, and S.N. Murphy, A clustering approach\n_for detecting implausible observation values in electronic health_\n_records data. BMC Med Inform Decis Mak, 2019. 19(1): p. 142._\n\n98. Ling, Y., et al. An error detecting and tagging framework for reducing\n_data entry errors in electronic medical records (EMR) system. 2013._\n\n99. Lai, K.H., et al., Automated misspelling detection and correction\n_in clinical free-text records. Journal of Biomedical Informatics,_\n2015. 55: p. 188–195.\n\n100. Peng, M., et al., Exploration of association rule mining for cod_ing consistency and completeness assessment in inpatient admin-_\n_istrative health data. Journal of Biomedical Informatics, 2018._\n**79(February): p. 41–47.**\n\n101. Wang, Z., M. Penning, and M. Zozus, Analysis of Anesthesia\n_Screens for Rule-Based Data Quality Assessment Opportunities._\nStud Health Technol Inform, 2019. 257: p. 473–478.\n\n102. Brown, J.S., M. Kahn, and D. Toh, Data quality assessment for\n_comparative effectiveness research in distributed data networks_\nMedical Care, 2013. 51(8 SUPPL.3): p. S22-S29.\n\n103. Johnson, S.G., et al., Application of an ontology for character_izing data quality for a secondary use of EHR data. Applied_\nClinical Informatics, 2016. 7(1): p. 69–88.\n\n104. Boselli, R., et al., A policy-based cleansing and integration\n_framework for labour and healthcare data. 2014. p. 141–168._\n\n105. Ferrao, J.C., et al., Preprocessing structured clinical data for\n_predictive modeling and decision support. A roadmap to tackle_\n_the challenges. Appl Clin Inform, 2016. 7(4): p. 1135–1153._\n\n106. Almeida, M., et al., Basic Formal Ontology 2.0 2015.\n\n**Publisher’s Note Springer Nature remains neutral with regard to**\njurisdictional claims in published maps and institutional affiliations.\n\n\n-----\n\n"