A generalized Data Quality Assessment Framework for Diverse Health Datasets with varied Contradiction Rules

A generalized Data Quality Assessment Framework for Diverse Health Datasets with varied Contradiction Rules | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A generalized Data Quality Assessment Framework for Diverse Health Datasets with varied Contradiction Rules Khalid O. YUSUF, Katharina JÖRẞ, Jendrik RICHTER, Sabine HANSS, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6929289/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Data quality assessment (DQA) in health research is guided by rules that express several indicators such as completeness, conformance, and plausibility. However, contradiction rules vary with diverse health domains. For example, while in cardiology, it is established that diastolic blood pressure should not exceed corresponding systolic measurements, in the biobank, it is an anomaly for the blood sample to remain stateless. Despite efforts to harmonize data quality indicators, implementations of DQA are often limited by predefined rules and number of evaluated interdependent items which makes tool reusability difficult. A generalized DQA framework that allows definition of custom contradiction rules compatible with datasets from varied health domains is yet to be reported. Objective The aim of this work is to develop a generalized DQA tool that can assess the quality of diverse health datasets, with validation focused on two distinct datasets of different structures and rules from the cardiology domain. Specifically, the validation datasets include 1) HiGHmed use case sensor data, and 2) the Medical Information Mart for Intensive Care - Electrocardiograms (MIMIC-IV-ECG) dataset. Methods A DQA tool that decentralizes assessment rule generation was developed in R. Among the elicited requirements for the tool development are: 1) support for custom rule definition, 2) support for diverse datasets and formats, 3) detailed and traceable assessment report, and 4) interactive interface to aid usability of the tool. To test support for diverse data formats, the tool incorporates both Fast Healthcare Interoperability Resources (FHIR) and open Electronic Health Record (openEHR) parsers in R that decomposes the bundled resources within the FHIR documents and openEHR compositions in the HiGHmed sensorik study. The generalizability of the DQA tool was tested using custom rules that are specific to the HIGHmed sensor data and the MIMIC-IV-ECG dataset. For usability of the DQA tool, a Shiny R interface was implemented for an interactive DQ assessment with choice datasets and user-defined rules. Results All elicited requirements including those desired by target users were fulfilled. The generalized framework has two segments including an overarching function that allows custom rule generation and an inner function that evaluates the defined rules on target dataset. Dataset from the HiGHmed use case cardiology were first evaluated and subsequently, the tool was successfully reused for the MIMIC-IV-ECG datasets. Also, the tool evaluates contradictions using custom rules defined by study personnel and imported study data in the interactive module. Discussion A reusable DQA tool is offered that analyzes health data sourced from different data models to produce comparable results across multiple study sites. The interactive DQA interface would assist study personnel in monitoring study data to address emerging data quality concerns. The usability of the DQA tool means that domain experts have the freedom to define and integrate rules tailored to their domain. Medical Informatics Data quality health data medical wearables FHIR openEHR Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Data quality (DQ) in health research relies on domain-specific rules that are transformed into executable Boolean logic in DQ systems. While some quality indicators like completeness can be assessed through generalized approaches, contradiction assessment presents unique challenges: contradiction logics operate in pairs (e.g., comparing home glucose measures against clinic readings), and rules range from basic logical checks (e.g. age disparity in baseline and follow-up visits) to complex empirical evaluations (e.g. inconsistencies in pre-analytic states of blood samples). Therefore, generic implementation of contradictory dependencies is hindered by its multi-deminsions. Several tools have been developed to support contradiction assessment in health research 1–6. However, the implementations face significant constraints due to the incompatibility of implemented rules with diverse DQ rules from different health domains 2,3,7. Though efforts have been made to standardize DQ indicators through harmonization, the reusability of the data quality assessment (DQA) tools for contradiction assessment is challenging 8,9. This is coupled with additional effort for metadata definition to guide the evaluation of predefined checks. Important parameters such as the flexibility in item definition that would support custom rule definition are missing in the existing implementations. Three DQA tools that support contradiction assessment were compared and following gaps were observed. While openCQA 2 suits the assessment of exported open Electronic Health Record (openEHR) compositions, it does not support the Fast Healthcare Interoperability Resources (FHIR) format of similar items because the rules expect an openEHR syntax with defined archetype-paths. DQA can only be realized using the tool if the FHIR resources are initially mapped to openEHR compositions. With ohdsi_DQD 3, source data models must first be mapped to the specified version of Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) and fitting rules must be incorporated in the metadata with expanded Structured Query Language (SQL) queries to run a successful DQA for new datasets. This limitation is compounded when the implemented rules are incompatible with intended use cases. While dataquieR 7 supports the import of structured datasets, FHIR document bundles and openEHR compositions require an initial transformation which is not supported by the tool. Also, separate rules compatible with new use cases must be predefined in the tool, necessitating expansion of interdependent items and ultimately the rebuild of significant part of the tool infrastructure. This encourages the implementation of multiple tools for varied rule specifications (e.g. basic vs. broader rules) as seen with dqGECCO tool 5. In this work, we propose a generalized approach that decentralizes the rule generation in the DQA tool. This is aimed at developing a reusable tool that accepts diverse user-defined rules compatible with varied health datasets. To determine the applicability of the generalized DQA framework, datasets with FHIR and openEHR structures from a multi-site HiGHmed use case cardiology sensorik study (referred to in this work as HiGHmed sensor data) were first analyzed. Subsequently, distinct contradiction rules for the Medical Information Mart for Intensive Care - Electrocardiograms (MIMIC-IV-ECG) were evaluated. While the HiGHmed use case cardiology sensorik study captures fitness data including steps and heart rates (HR) through the Apple Watches warn by study participants, the MIMIC-IV-ECG dataset is an open source diagnostic records derived from the 12-leads electrocardiograms (ECG) of patients 10,11. The HiGHmed sensor data collection is structured such that the bulk of the continuous measurements are bundled as FHIR documents and transmitted through the health App to designated FHIR endpoints of the participating Medical Data Integration Centres (MeDICs). While some of the study sites retained the FHIR format, other sites bridged their endpoint to receive data in openEHR format. A recent study identified the lack of DQA tools to support FHIR bundles within the Medical Informatics Initiative (MII) 12. Also. previous studies have reported measurement inconsistencies and compliance issues as some of the quality concerns that must be addressed regarding the use of sensor data 13, 14. Therefore, this use case presents an opportunity to utilize the generalized DQA tool for the assessment of sensor data with FHIR and openEHR structures. Furthermore, the MIMIC-IV-ECG dataset has been widely used in building models for predicting various cardiac conditions such as myocardial infarction and rhythm abnormalities 15. However, it is important to determine the consistency of the cardiac annotations and the machine measurements that contribute to the accuracy of the model classifications 16. Considering that the MIMIC-IV-ECG dataset has distinct structure and rules from the HiGHmed sensor data, it allows us to assess the adaptability of the generalized DQA tool to new health data environments. Objective In this work, our goal is to develop a generalized DQA tool that is compatible with varied contradiction rules applicable to diverse health datasets, with validation focused on two distinct datasets from the cardiology domain. Specifically, the two structurally distinct datasets include: 1) the HiGHmed sensor data, and 2) MIMIC-IV-ECG dataset. The objective of this approach is to bring flexibility into DQA tool implementation such that it can be extended to diverse health data contexts (e.g. biobanking). 2. Methods 2.1. Diverse Data Collections 2.1.1. HiGHmed Use Case Cardiology Sensorik Study The HiGHmed use case cardiology sensorik study is a project within the MII in which novel health-IT solutions aimed at increasing access to patients’ medical data for research are being developed ADDIN10. NThis study utilizes the Apple Watch to continuously record fitness metrics, specifically steps and heart rates (HR) 17,18. The integrated health application employs FHIR profiles as means of data exchange with the participating MeDICs. While two MeDICs (Göttingen and Würzburg) receive bundled resources as FHIR binaries, two other MeDICs (Cottbus and Hannover) bridged the FHIR endpoint to map the resources into openEHR compositions 19. The Kansas City Cardiomyopathy Questionnaire (KCCQ) was also documented to assess the health status of patients 20. Daily survey of body weight, resting HR and blood pressure (BP) were measured manually. At the Göttingen data collection site, six participants were onboarded while Würzburg enrolled about 50 participants. Cottbus and Hannover have two and nine study participants respectively. All MeDICs of the partner sites transferred either FHIR binaries or archetype query language (AQL)-resultset to Göttingen for a centralized data quality analysis. As depicted in Fig. 1 , the base64 encoded FHIR binaries in JSON format bundles a collection of FHIR documents containing the resources. The binaries were transformed and decomposed into separate FHIR documents using PowerShell script. All relevant resources were extracted from the documents and loaded into R using parse_fhir function linked to the DQ analysis tool. About 13,102 document bundles were extracted from Würzburg FHIR binaries with each document bundle storing up to 600 observations. This resulted in a total of 138,637 step measurements, 497,923 device HR measurements, 32 KCCQ responses, and 6,693 daily survey responses including manual HR, systolic and diastolic BP, and weight for 38 subjects in the analyzed dataset. Cottbus transmitted 509 AQL-resultset for 2 subjects and a total of 26,293 step, 113,063 HR observations, 11 KCCQ responses, and 840 survey were extracted using parse_openehr function. In Göttingen, the decomposed FHIR resources from 1,553 documents included 152,391 device HR, 70,204 steps, 20 KCCQ responses, and 1,559 daily survey. Hannover transferred preprocessed openEHR composition with 181,148 heart rate measurements, 49,139 steps, 476 daily surveys, and 8 KCCQ responses. For our analysis, we considered three contradictory dependencies including: 1) conflicting device HR measurements, 2) inconsistencies in device and manual HR measurements, and 3) conflicting responses in KCCQ. 2.1.2. MIMIC- IV-ECG Dataset MIMIC-IV-ECG is an open source ECG diagnostic dataset with over 800,000 records of 160,000 patients documenting cardiology reports and associated machine-generated measurements 11. The waveform_note_links.csv and report.csv stored the metadata of the ECG records and cardiologist reports, such as study identifiers, subject identifiers, time of ECGs, note identifiers of cardiologist reports, International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) diagnosis codes, and nametags for the ECG files. In the machine_measurements.csv, a summary report of machine-generated evaluations, such as heart rhythm analysis and HR, was documented. While the dataset supports researchers in automating the evaluation of cardiac-related conditions 15, there are concerns about the consistency of the ICD annotations documented during diagnosis and the machine-generated measurements 16. For our evaluation in this work, we considered the ICD annotations of all cardiac electronic devices including implanted pacemakers and defibrillators captured in the MIMIC-IV-ECG dataset as well as the corresponding pacing rhythms in the ECG measurements estimated through the machine algorithm. The detailed ICD codes are presented in Table 1 . The consistency evaluations target two cases including: 1) ICD annotation of pacemakers and defibrillators (Z95.0, Z95.810, Z45.01, Z45.02, & Z45.09) that are linked to hospital stay (has_stay_id) with corresponding machine pacing rhythms, and 2) machine pacing rhythms without ICD codes. Table 1 Description of International Classification of Diseases (ICD) annotation of pacemaker, defibrillator, and other cardiac devices. ICD-10 Code Description Z95.0 presence of cardiac pacemaker Z95.810 presence of automatic implantable cardiac defibrillator Z45.01 encounter for adjustment and management of cardiac pacemaker Z45.02 encounter for adjustment and management of automatic implantable cardiac defibrillator Z45.09 encounter for adjustment and management of other cardiac devices 2.2. Design of Generalized DQA Framework The design of the generalized DQA (gDQA) framework depicted in Fig. 2 uses higher-order function in R programming that creates and returns another function as output 21. According to the documented requirements in Table 2 , the tool requires an overarching function to fulfill ER1 (elicited requirement) that accepts custom rules generated externally, and an inner function (ER2) that evaluates the rules on the target dataset. As shown in Fig. 2 , although the domain rules are plugged into the main tool with a call during execution, these rules are generated as an independent object using the rule generator function. This approach is intended to allow diverse health domains to define and integrate applicable contradiction rules suitable for the study data to be analyzed. This is a departure from current practice in existing tools, where rules and acceptable variables are fixed in the tools without the flexibility to define custom rules applicable to varied use cases. Also, apart from the user-defined rules, there are no additional required metadata to manually guide the operation of the tool on what rules to evaluate on selected items 2,5. In compliance with ER3 in Table 2 , all rules were created at the atomic level and labeled appropriately to make the identified anomalies traceable. The generated contradiction rules are evaluated on each row of the study data, with every violation cached and returned as output. Arguments in the rule generator include labels of the rule-set to be created and list of rule expressions defined within it. During execution, users select the respective labels while the DQA tool imports the rules listed within the selected label. The output of the rule generator serves as the inner function that evaluates the parsed study data to produce an assessment report. To enhance the usability of the tool as required in ER4, an interactive module is created in Shiny R. The module allows import of datasets of different formats including csv, openEHR, and FHIR document bundles as required in ER2. Users are guided with the Help function in defining desired custom rules according to ER1. Through the engagement of prospective users of the tool including data managers and study nurses, the requirements ER5 addresses users’ preference for selection of specific custom rules from the created pool. This will streamline the assessment report to specific violations. Table 2 Requirement elicitation (ER) for the generalized data quality framework. Requirements Description Priority ER1 custom rule generator Required ER2 diverse datasets import (csv, fhir bundle, openEHR composition) Required ER3 detailed and traceable output Required ER4 usability support (interactive module) Required ER5 specific custom rule selection Desired 2.3. Contradiction Assessment in HiGHmed Sensor Data We analyzed contradictions in the decomposed resources from FHIR binaries and similar items from openEHR compositions in the HiGHmed sensor data using the gDQA tool. The implemented rules addressed three contradictions including 1) conflicting measurements within heart rates (HR) measured using the device, 2) inconsistencies in device and manual HR measurements, and 3) conflicting responses in KCCQ. To evaluate the contradictions in HR measured using the device, the effective time of each measurement is considered. The time difference between the preceding and the successive measurements is derived. A time difference of zero seconds is considered implausible provided the affected HR measurements are not duplicated. This would result in contradictory measurements recorded at the same time-point, which could bias the average HR of the subject for the day. Also, for the contradictions between device-recorded HR and those measured manually, the last HR measured by the device at the nearest time to the time of the manually captured HR was compared. The third check examines the responses by the study participants to the KCCQ questionnaire items. An example of a conflicting response is when the subject chooses to be satisfied and unsatisfied with a question regarding the quality of life on the same response date. The overall implication of such contradictions is that they affect the derived total score of the subject's health status. 2.4. Consistency Assessment in MIMIC-IV-ECG Dataset We implemented two consistency rules in the MIMIC-IV-ECG dataset focusing on the correlations between ICD code annotations of pacemakers and defibrillators against the machine-detected pacing rhythms: 1) annotation-to-measurement consistency, and 2) measurement-to-annotation consistency. For rule 1), we defined Boolean consistency rule to check whether patients with ICD codes associated with cardiac electronic devices (specifically pacemakers and defibrillators) show corresponding evidence of pacing rhythm in their ECG measurements. The implemented Boolean rule 2) evaluates cases where the ECG measurements indicate pacing rhythm but lacked corresponding ICD documentation for implanted cardiac devices. The inclusion criteria in both cases is the presence of has_stay_id which indicates whether the ECG records are associated to patients’ hospital stay. We estimated the total pacing records from the MIMIC-IV-ECG dataset as a sum of records that satisfy rule 1) and records that violates rule 2). The proportion of satisfactions or violations are determined by their frequency in the total pacing records. 3. Results 3.1. Implementation All the elicited requirements (ER) in Table 2 with the required and desired priority tags were implemented. The implementation source code can be found in the Gitlab repository of the project 22. The tool supports both predefined and custom rules as required by ER1. Users are not limited by the number of evaluated dependencies within items since the rule construct uses R extract operator ( $ ) which can link all items within a name list. Users have the option to parse datasets in three data formats including csv, FHIR, and openEHR for analysis based on ER2. The parse_fhir and parse_openehr functions in the background extracts the resources within the FHIR document bundle and openEHR compositions and transform them into tabular form. A preview of the decomposed resources is enabled as shown in the interface in Fig. 3 . ER3 was fulfilled by providing separate fields for users to enter labels for their custom rules and specify the rule conditions. A rule_processor function converts the user-defined rules into expressions that are executable in R. After running the analysis, the labels of each contradiction rule are used to describe violated rules in the assessment report. Figure 3 shows the interface of the interactive module as required in ER4 with implemented features to support users without programming experience in fulfilling ER1, ER2, ER3, and ER5. 3.2. Conflicting Device-Recorded Heart Rate Measurements The analysis of simulteneously-recorded heart rates (HR) using the wearable device revealed measurement conflicts across all study sites. At the Würzburg site, 9,968 (2.1%) of 497,933 observations contained conflicting records as shown in Table 3 . The measurement intervals averaged 428 seconds, excluding outliers (0 seconds and 2,689,109). Zero-second epochs represented conflicting values in successive measurements while the maximum epoch resulted from prolonged device outage. The Göttingen site demonstrated fewer conflicts, with 250 (0.16%) conflicting observations among 152,391 HR measurements. The epoch distribution showed a minimum of 0 seconds, first quartile at 186 seconds, median of 289 seconds, mean of 310 seconds, third quartile at 359 seconds, and maximum of 69,121 seconds. This distribution indicates HR averaging windows predominantly fell between 186–359 seconds, consistent with the time path visualization in Fig. 4 . At the Hannover site, 140 conflicting measurements were identified in 118,148 HR readings. The Cottbus site showed the lowest rate with only 54 (0.05%) conflicting observations among 113,063 measurements. These conflicting values represent measurement discrepancies rather than simple duplicates, as HR values at identical time-points differ. Although all study sites demonstrated high plausibility rates as shown in Table 3 , these conflicting measurements could potentially bias daily HR averages. While the Apple Watch system acknowledges potential sensor limitations that may produce anomalous readings, our findings confirm that such measurement conflicts were minimal across all study sites 17. Table 3 Plausibility rate for heart rate (HR) measurements Study site Total HR Observations Plausible Records Implausible Records Plausible Rate (%) Würzburg 497923 487955 9968 97.99 Göttingen 152391 152141 250 99.84 Cottbus 113063 113009 54 99.95 Hannover 181148 181008 140 99.92 3.3. Inconsistent Manual and Device-Recorded Heart Rates A total of 2,363 observations were collected across all study sites, representing paired manual and device-recorded HR from 47 study participants. As presented in Table 4 , analysis of these paired measurements showed varied patterns of agreement and discrepancy. The majority of the comparison revealed 1,186 device-recorded observations exceeding the manual measurements by less than 20 bpm. Conversely, in 853 observations, the manual measurements exceeded the device readings by 20 bpm. Perfect agreement between the manual and device measurements occurred in 182 observations. Notable discrepancies exceeding 20 bpm were observed in 142 observations, with the device measurements exceeding manual readings in 96 instances and manual measurements exceeding device readings in 46 instances. As illustrated in Fig. 5 , while most discrepancies observed are within 20 bpm, the presence of larger discrepancies in about 6% of the observations deserve attention. These substantial differences could potentially trigger unnecessary medical interventions if not properly verified with contextual evidence. Additionally such discrepancies visible to users through health applications may unnecessarily elevate anxiety regarding health status. Table 4 Breakdown of heart rate (HR) measurement difference between device and manual methods. Total HR Readings Equal Measurement Device HR > 20bpm Manual HR > 20bpm Device HR < 20bpm Manual HR < 20bpm 2363 182 96 46 1186 853 3.4. Conflicting Responses to Cardiology Questionnaire From 47 subjects, conflicting responses were observed in two subjects (4.3%). Figure 6 illustrates the pattern for one representative subject who provided contradictory responses on the same study day. The conflicts were observed in two distinct KCCQ items: 1) subject’s ability to run fast or hurry up, and 2) subject’s quality of life. The subject simultaneously on the same study day indicated being “hindered a little” and “moderately hindered” while hurrying up or running fast. For the question about satisfaction with living the remainder of life with heart failure, the same subject submitted both “satisfied” and “unsatisfied” responses on the same day. These conflicting responses compromise the accuracy of the derived overall health status score, which is calculated from the KCCQ responses. Such inconsistencies introduce uncertainty in the assessment of the subject’s true health status and could potentially lead to inadequate clinical decisions if not verified. According to the study protocol, subjects were required to complete the KCCQ at scheduled time-points: on the first day of enrollment, three times at four-week intervals, and subsequently at three-month intervals. The observed conflicts reveal limitations in the data collection, specifically the lack of controls in the health application to prevent or flag multiple submissions on a single response date. 3.5 Consistency of Cardiac Annotation and Machine-Detected Rhythm Our analysis revealed that only 26,990 records out of a total of 800,035 records in the MIMIC-IV-ECG dataset have pacing detection. As presented in Fig. 7 , among these 26,990 records with any pacing detection, 20,652 records show consistent match between ICD annotation and machine-detected pacing rhythm. The remaining 6,338 records were cases where there are machine-pacing rhythm without corresponding ICD codes for the cardiac devices. The implication of this annotation gap is that ICD codes alone are not sufficient to determine the presence of pacemakers or defibrillators. Model reliance on solely ICD codes as ground truth may lead to missing actual pacing cases. 4. Discussion Principal Contributions Our research introduces gDQA , a reusable DQA framework that represents a significant shift from the constraints of predefined rules prevalent in existing contradiction assessment tools. By designing a flexible rule parsing system, gDQA accepts custom contradiction rules while maintaining a consistent assessment. By evaluating gDQA with distinct health datasets (HiGHmed sensor data and MIMIC-IV-ECG) of different structures and contradiction rule contexts, we demonstrated the adaptability of the tool for diverse health datasets and rules. This validation is particularly relevant considering the challenge of maintaining a consistent DQA in diverse health data environments. Though contradiction rules in different health domains vary however, the assessment methods implemented through gDQA complement the established standardized indicators proposed by Kahn et al. and Schmidt et al. 6,7. gDQA extends the utility of the harmonized frameworks by providing a flexible implementation. With the decentralization of the assessment rules parsed into the tool, the generalized framework supports researchers to define and integrate domain-specific quality checks while maintaining an alignment with established DQ standards. The applicability of the tool for datasets of varied formats is an evidence that our approach supports multi-site collaborations. This is particularly relevant in an era where data sharing across health disciplines transcends institutional and national boundaries 23. Furthermore, in clinical trial settings, where timely quality assessment is essential for monitoring data integrity throughout the data collection process, gDQA offers particular advantage. The framework’s support for custom rules enables trial-specific quality checks to be implemented rapidly, facilitating near real-time assessment report 24. Data Quality Issues in Medical Wearables The gDQA tool identified three contradiction patterns in the sensor datasets. The inconsistencies in HR measurements when using a device against the manual method can lead to confusion in treatment options or unnecessary intervention by the subjects to address sudden spike. While it is understandable that the device returns an estimated average of HR measurements over a scheduled interval, two different estimates at the same time-point from the same device are controversial and require further investigation. In addition, responses to predefined cardiology questionnaires serve a crucial purpose in determining subjects' health status at different activity levels. Therefore, providing an accurate response at the scheduled interval is important. As observed in the findings, the few reported gaps in the questionnaire were due to unrestricted data entry outside the scheduled response interval. Inconsistency in ICD Annotation and ECG Machine Measurements The inconsistencies observed in about 23% of the pacing records is an opportunity to improve the validation of ICD annotations in real-world applications. However, these cases may also be linked to undocumented temporary pacing devices during the patients’ hospital stay or devices implanted at different hospital not captured in the evaluated medical records. An expert review of these findings would confirm if they are true pacing or false positives from the machine algorithm. For model development, it is encouraged to consider both ICD codes and machine measurements as complementary signals. gDQA and Existing Tools A common gap identified in existing DQA frameworks is that assessment rules are predefined within the tools with extra work required for metadata definition to guide the evaluation of already embedded rules, which is time-consuming 2,5. This has been addressed by reducing the attributes of the gDQA tool to user-prepared study data and custom DQA rules. Users are aided with the populated item list to select from when defining rules to minimize manual efforts. Our approach shows that rather than modifying DQA tools to incorporate new domain rules, only the rule list is expanded as desired. This study has demonstrated that similar to the harmonization of data quality indicators, the tool implementing the indicators can equally be generalized. The findings in this study revealed contradictions within isolated items which Cho et al. 13 classified as an atemporal plausibility of surrounding measurements. However, while the averaging time window influenced the contradictory HR findings, the conflicting questionnaire responses (also isolated) were illogical owing to multiple data entry on same response date. Therefore, when the event times of the conflicting findings are compared without any reference to the actual measurements or responses, they already qualify for duplicated entries, which is another DQ indicator. The actual measurements and responses were only required to validate the contradictions. Hence, these findings are simply contextual with the measurement time. This is an indication that as timestamps provide context to rule out contradictions within interdependent items, isolated or interdependent items can also be contextual to timestamps 25. Limitations While efforts have been put into making the gDQA tool user friendly, data preparation involving patient records distributed across multiple tables requires varied levels of expertise among potential users. Additionally, to reduce the reliance on domain expertise for effective rule definition, the current implementation will benefit from a large collection of established assessment rules that cut across different health domains. This can only be achieved through collaborations with domain experts. In future work, we envisage that rule generation can be automated based on the characteristics of the items in datasets and domain ontologies to limit human efforts in rule processing. Further evaluations performed in diverse clinical settings (e.g. support for DICOM, genomic datasets) will strengthen the performance of the tool by identifying potential improvements. Conclusion The development of a reusable DQA tool that is compatible with multiple health datasets is a significant contribution in health data management. The agnostic nature of the tool to distinct rules as tested on sensor data from medical wearables, and the MIMIC-IV-ECG dataset demonstrates its adaptability. Also, the freedom of definition of assessment rules empowers domain experts to define effective rules on ad-hoc basis while maintaining established DQ standards without being limited by predefined rules. Declarations Acknowledgements: The work is jointly funded by MII, NUM, grant number 01KX2121 and DZHK, grant number 81Z0300108 Conflict of Interest: The authors declare that there is no conflict of interest. Ethical approval and consent: Use of HiGHmed sensor data falls under the Medical Informatics Initiative (MII) data quality assurance covered by the ethics committee votes (09.10.2020) for the project. References Mariño J, Kasbohm E, Struckmann S, Kapsner LA, Schmidt CO. R Packages for Data Quality Assessments and Data Monitoring: A Software Scoping Review with Recommendations for Future Developments. Applied Sciences. 2022 Apr 22;12(9):4238,DOI:10.3390/app12094238. Tute E, Scheffner I, Marschollek M. A method for interoperable knowledge-based data quality assessment. BMC Med Inform Decis Mak. 2021 Dec;21(1):93. Blacketer C, Defalco FJ, Ryan PB, Rijnbeek PR. Increasing trust in real-world evidence through evaluation of observational data quality. J Am Med Inform Assoc. 2021 Sep 18;28(10):2251–7. Yusuf K, Tahar K, Sax U, Hoffmann W, Krefting D. Assessment of the Consistency of Categorical Features Within the DZHK Biobanking Basic Set. In: Röhrig R, Grabe N, Hoffmann VS, Hübner U, König J, Sax U, et al., editors. Studies in Health Technology and Informatics [Internet]. IOS Press; 2022 [cited 2022 Sep 10]. Available from: https://ebooks.iospress.nl/doi/10.3233/SHTI220809. Yusuf KO, Miljukov O, Hanß S, Schoneberg A, Wiesenfeldt M, Stecher M, et al. Consistency as a Data Quality Measure for German Corona Consensus items mapped from National Pandemic Cohort Network data collections. Methods Inf Med. 2023 Jan 3;a-2006-1086. Johnson SG, Pruinelli L, Hoff A, Kumar V, Simon GJ, Steinbach M, et al. A Framework for Visualizing Data Quality for Predictive Models and Clinical Quality Measures. AMIA Jt Summits Transl Sci Proc. 2019;2019:630–8. Struckmann S, Mariño J, Kasbohm E, Salogni E, Schmidt CO. dataquieR 2: An updated R package for FAIR data quality assessments in observational studies and electronic health record data. Journal of Open Source Software. 2024 Jun 28;9(98):6581. Kahn MG, Callahan TJ, Barnard J, Bauck AE, Brown J, Davidson BN, et al. A Harmonized Data Quality Assessment Terminology and Framework for the Secondary Use of Electronic Health Record Data. EGEMS (Wash DC). 2016 Sep 11;4(1):1244. Schmidt CO, Struckmann S, Enzenbach C, Reineke A, Stausberg J, Damerow S, et al. Facilitating harmonized data quality assessments. A data quality framework for observational health research data collections with software implementations in R. BMC Med Res Methodol. 2021 Dec;21(1):63. Haarbrandt B, Schreiweis B, Rey S, Sax U, Scheithauer S, Rienhoff O, et al. HiGHmed - An Open Platform Approach to Enhance Care and Research across Institutional Boundaries. Methods Inf Med. 2018;57(S 01):e66–81. Gow B, Pollard T, Nathanson LA, Johnson A, Moody B, Fernandes C, et al. MIMIC-IV-ECG: Diagnostic Electrocardiogram Matched Subset [Internet]. PhysioNet; [cited 2025 May 8]. Available from: https://physionet.org/content/mimic-iv-ecg/1.0/ Draeger C, Tute E, Schmidt CO, Waltemath D, Boeker M, Winter A, et al. Identifying Relevant FHIR Elements for Data Quality Assessment in the German Core Data Set. In: Caring is Sharing – Exploiting the Value in Data for Health and Innovation [Internet]. IOS Press; 2023 [cited 2025 Apr 18]. p. 272–6. Available from: https://ebooks.iospress.nl/doi/10.3233/SHTI230117 Cho S, Weng C, Kahn MG, Natarajan K. Identifying Data Quality Dimensions for Person-Generated Wearable Device Data: Multi-Method Study. JMIR Mhealth Uhealth. 2021 Dec 23;9(12):e31618. Purta R, Hachen D, Striegel A, Poellabauer C, Lizardo O, et al. Exploring Compliance: Observations from a Large Scale Fitbit Study. In: Proceedings of the 2nd International Workshop on Social Sensing [Internet]. Pittsburgh PA USA: ACM; 2017 [cited 2024 May 3]. p. 55–60. Available from: https://dl.acm.org/doi/10.1145/3055601.3055608. Strodthoff N, Lopez Alcaraz JM, Haverkamp W. Prospects for artificial intelligence-enhanced electrocardiogram as a unified screening tool for cardiac and non-cardiac conditions: an explorative study in emergency care. European Heart Journal - Digital Health. 2024 Jul 1;5(4):454–60. McKeen K, Oliva L, Masood S, Toma A, Rubin B, Wang B. ECG-FM: An Open Electrocardiogram Foundation Model [Internet]. arXiv; 2024 [cited 2025 May 9]. Available from: http://arxiv.org/abs/2408.05178. Apple Support [Internet]. 2023 [cited 2024 Mar 12]. Monitor your heart rate with Apple Watch. Available from: https://support.apple.com/en-us/HT204666. Apple Support [Internet]. [cited 2024 May 3]. Track daily activity with Apple Watch. Available from: https://support.apple.com/guide/watch/track-daily-activity-with-apple-watch-apd3bf6d85a6/watchos. FHIR-Bridge. https://github.com/NUM-Forschungsdatenplattform/num-fhir-bridge Spertus JA, Jones PG, Sandhu AT, Arnold SV. Interpreting the Kansas City Cardiomyopathy Questionnaire in Clinical Trials and Clinical Care: JACC State-of-the-Art Review. Journal of the American College of Cardiology. 2020 Nov 17;76(20):2379–90. Wickham H. Introduction | Advanced R [Internet]. [cited 2025 Apr 17]. Available from: https://adv-r.hadley.nz/fp.html. Source code for generalizedDQA development: https://gitlab.gwdg.de/medinfpub/kvf-group/generalized-dq-tool Health Data Space (EHDS) [Internet]. [cited 2024 Nov 26]. Available from: https://www.european-health-data-space.com/. Houston L, Probst Y, Yu P, Martin A. Exploring Data Quality Management within Clinical Trials. Appl Clin Inform. 2018 Jan;9(1):72–81. Yusuf KO, Chaplinskaya-Sobol I, Schoneberg A, Hanss S, Valentin H, Lorenz-Depiereux B, et al. Impact of Clinical Study Implementation on Data Quality Assessments - Using Contradictions within Interdependent Health Data Items as a Pilot Indicator. Stud Health Technol Inform. 2023 Sep 12;307:152–8. Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6929289","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":473506295,"identity":"42644c08-dc6b-4c3d-926c-c55b087dde7b","order_by":0,"name":"Khalid O. YUSUF","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABjUlEQVRIie2SzUrDQBDHZ1lILqu9Jmh9hpGFSGnRZwkBe2lRECSo1JVCvCi9tiD4CgUv3kwJ6CXgNVAPFi8eekiph4r4sZukxUL1LJj/YXdmmd/ODDMAuXL9Renq2AVEoMJnyRMZPAJQZYg0hqbXnIsKIVOEcvyOGOwXBDLEMmY/JwjMIQXKeBzjJt/Qeyf+EAL7su1oB5FbBuw3BXlxK40tfSl4Hl1DsZUgZpNZZhsdq3Rmi96FRLqRo/VrYRXwoSfoargtC1veLnVC4J0kDQbMogxpBX1bBEwhxs5tv+4FgJF9+mZ6gUSYxZc8sLt+hvAxw+MK3g9SRBW2V/9IEEFN7zNF3iVyM0VwhWFgqYAEEbIwWhcpQkaerxD+RFSWrBdtX/ZyxzEayF6wyrvhE12p3VaZqbJA6Jhnsnhy7hm8nWQp6M2rOHYP17v3ThAP3XLx8tQh49pReW05qj6SibtZ0Fshj1+9SrElFuwCNmdmMg2azUSTkzIWbg9AY94lk2zg8Q/xuXLlyvUf9AVu+pJiUx4AJwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0003-2556-6898","institution":"Department of Medical Informatics, University Medical Centre Göttingen, Germany","correspondingAuthor":true,"prefix":"","firstName":"Khalid","middleName":"O.","lastName":"YUSUF","suffix":""},{"id":473506296,"identity":"8665b68c-0b8e-4bd2-8f2e-8ef0f58267fc","order_by":1,"name":"Katharina JÖRẞ","email":"","orcid":"","institution":"Department of Medical Informatics, University Medical Centre Göttingen, Germany","correspondingAuthor":false,"prefix":"","firstName":"Katharina","middleName":"","lastName":"JÖRẞ","suffix":""},{"id":473506297,"identity":"c420b0dc-85b8-426b-b501-9e202fab212c","order_by":2,"name":"Jendrik RICHTER","email":"","orcid":"","institution":"Department of Medical Informatics, University Medical Centre Göttingen, Germany","correspondingAuthor":false,"prefix":"","firstName":"Jendrik","middleName":"","lastName":"RICHTER","suffix":""},{"id":473506298,"identity":"3388b97f-c48e-47e3-9ecb-53613b75416e","order_by":3,"name":"Sabine HANSS","email":"","orcid":"","institution":"Department of Medical Informatics, University Medical Centre Göttingen, Germany","correspondingAuthor":false,"prefix":"","firstName":"Sabine","middleName":"","lastName":"HANSS","suffix":""},{"id":473506299,"identity":"2ba4d1fe-0eb5-4d94-8348-1ceee86ba0d0","order_by":4,"name":"Irina CHAPLINSKKAYA-SOBOL","email":"","orcid":"","institution":"Department of Medical Informatics, University Medical Centre Göttingen, Germany","correspondingAuthor":false,"prefix":"","firstName":"Irina","middleName":"","lastName":"CHAPLINSKKAYA-SOBOL","suffix":""},{"id":473506300,"identity":"d54600d6-10ef-42c9-9521-2dd2832b8894","order_by":5,"name":"Robert KOSSEN","email":"","orcid":"","institution":"Department of Medical Informatics, University Medical Centre Göttingen, Germany","correspondingAuthor":false,"prefix":"","firstName":"Robert","middleName":"","lastName":"KOSSEN","suffix":""},{"id":473506301,"identity":"8a52697c-fa43-4c68-ade0-495e29b4ddfb","order_by":6,"name":"Lennart GRAF","email":"","orcid":"","institution":"Department of Medical Informatics, University Medical Centre Göttingen, Germany","correspondingAuthor":false,"prefix":"","firstName":"Lennart","middleName":"","lastName":"GRAF","suffix":""},{"id":473506302,"identity":"1b2f3540-b87a-400c-9a13-20b6815e7619","order_by":7,"name":"Dagmar KREFTING","email":"","orcid":"","institution":"Department of Medical Informatics, University Medical Centre Göttingen, Germany","correspondingAuthor":false,"prefix":"","firstName":"Dagmar","middleName":"","lastName":"KREFTING","suffix":""}],"badges":[],"createdAt":"2025-06-19 08:53:22","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-6929289/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6929289/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":85186105,"identity":"51c30172-26da-47a0-a6de-0a711cfd0f83","added_by":"auto","created_at":"2025-06-23 08:09:21","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":113969,"visible":true,"origin":"","legend":"\u003cp\u003eHiGHmed Sensor Data Extract, Transform, and Loading Workflow. Sensor data from Apple-watch are bundled as FHIR binaries and transfered to the FHIR endpoints of participating MeDICs. MeDIC - medical data integration centre. Two MeDICs receive FHIR binaries directly while two other MeDICs bridged the FHIR endpoint to receive openEHR compositions. FHIR binaries and archetype result-set are decomposed before loading.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6929289/v1/e33fd0c9e4198dde6f3d6674.png"},{"id":85186108,"identity":"7c652c5d-fef2-4fb3-bf39-535fb076206e","added_by":"auto","created_at":"2025-06-23 08:09:21","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":263685,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow of the generalized data quality assessment (gDQA) framework showing integration of domain rule generator and parsers for FHIR document bundles and openEHR compositions. csv - comma seperated values, FHIR - fast healthcare interoperability resources, openEHR - open electronic health records.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6929289/v1/14b328705bf57df6545290ee.png"},{"id":85186672,"identity":"de69e713-466c-4e3c-b34d-e164bcfc1649","added_by":"auto","created_at":"2025-06-23 08:17:21","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":199714,"visible":true,"origin":"","legend":"\u003cp\u003eInteractive Interface for Data Quality Assessment that allows users to define custom rules for internal use and import choice datasets (csv, FHIR, and openEHR formats).\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6929289/v1/cf27a3c63ef1009cd5f75182.png"},{"id":85186106,"identity":"72569382-0f16-44a5-bf9a-25bbc331796c","added_by":"auto","created_at":"2025-06-23 08:09:21","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":385077,"visible":true,"origin":"","legend":"\u003cp\u003eDual chart showing the time difference between successive heart rate (HR) measurements and the actual difference in conflicting device-generated HR. Timelag 0 (in red) means two measurements were recorded at the same time-point.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6929289/v1/5a95e8793b80599a47f8a8c7.png"},{"id":85186115,"identity":"6fc7e3bd-83fe-46fd-ba71-8701f3ceb2ed","added_by":"auto","created_at":"2025-06-23 08:09:21","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":354675,"visible":true,"origin":"","legend":"\u003cp\u003eInconsistent heart rates (HR) measurements between the device and manual methods.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6929289/v1/9d258705174c63a0d6c2395e.png"},{"id":85186123,"identity":"3a8a1b9b-1304-4926-8586-402d55a8dda7","added_by":"auto","created_at":"2025-06-23 08:09:21","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":379254,"visible":true,"origin":"","legend":"\u003cp\u003eResponses provided by study participants to the Kansas City Cardiomyopathy Questionnaire (KCCQ). Conflicts are observed in answers to questions Q1c (in green) and Q7 (blue) on same day.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-6929289/v1/4648057f011b238104a042d4.png"},{"id":85186118,"identity":"6cd11710-3925-402d-a5b9-cbc5dbcb8efe","added_by":"auto","created_at":"2025-06-23 08:09:21","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":81855,"visible":true,"origin":"","legend":"\u003cp\u003eConsistency in pacing annotations with majority of the records (in blue) showing matches between ICD codes and machine-detected measurements. The inconsistent records (in red) are cases of machine measurements without corresponding ICD codes.\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-6929289/v1/9074ae9167f75fe894049460.png"},{"id":85188393,"identity":"417c54e6-91da-40e7-9a59-9ca04cd05a23","added_by":"auto","created_at":"2025-06-23 08:33:21","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2541826,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6929289/v1/aadd19d0-7e1f-4d14-a981-193eabc82fc8.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eA generalized Data Quality Assessment Framework for Diverse Health Datasets with varied Contradiction Rules\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eData quality (DQ) in health research relies on domain-specific rules that are transformed into executable Boolean logic in DQ systems. While some quality indicators like completeness can be assessed through generalized approaches, contradiction assessment presents unique challenges: contradiction logics operate in pairs (e.g., comparing home glucose measures against clinic readings), and rules range from basic logical checks (e.g. age disparity in baseline and follow-up visits) to complex empirical evaluations (e.g. inconsistencies in pre-analytic states of blood samples). Therefore, generic implementation of contradictory dependencies is hindered by its multi-deminsions. Several tools have been developed to support contradiction assessment in health research 1\u0026ndash;6. However, the implementations face significant constraints due to the incompatibility of implemented rules with diverse DQ rules from different health domains 2,3,7. Though efforts have been made to standardize DQ indicators through harmonization, the reusability of the data quality assessment (DQA) tools for contradiction assessment is challenging 8,9. This is coupled with additional effort for metadata definition to guide the evaluation of predefined checks. Important parameters such as the flexibility in item definition that would support custom rule definition are missing in the existing implementations. Three DQA tools that support contradiction assessment were compared and following gaps were observed. While \u003cem\u003eopenCQA\u003c/em\u003e 2 suits the assessment of exported open Electronic Health Record (openEHR) compositions, it does not support the Fast Healthcare Interoperability Resources (FHIR) format of similar items because the rules expect an openEHR syntax with defined archetype-paths. DQA can only be realized using the tool if the FHIR resources are initially mapped to openEHR compositions. With \u003cem\u003eohdsi_DQD\u003c/em\u003e 3, source data models must first be mapped to the specified version of Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) and fitting rules must be incorporated in the metadata with expanded Structured Query Language (SQL) queries to run a successful DQA for new datasets. This limitation is compounded when the implemented rules are incompatible with intended use cases. While \u003cem\u003edataquieR\u003c/em\u003e 7 supports the import of structured datasets, FHIR document bundles and openEHR compositions require an initial transformation which is not supported by the tool. Also, separate rules compatible with new use cases must be predefined in the tool, necessitating expansion of interdependent items and ultimately the rebuild of significant part of the tool infrastructure. This encourages the implementation of multiple tools for varied rule specifications (e.g. basic vs. broader rules) as seen with \u003cem\u003edqGECCO tool\u003c/em\u003e 5. In this work, we propose a generalized approach that decentralizes the rule generation in the DQA tool. This is aimed at developing a reusable tool that accepts diverse user-defined rules compatible with varied health datasets.\u003c/p\u003e \u003cp\u003eTo determine the applicability of the generalized DQA framework, datasets with FHIR and openEHR structures from a multi-site HiGHmed use case cardiology sensorik study (referred to in this work as HiGHmed sensor data) were first analyzed. Subsequently, distinct contradiction rules for the Medical Information Mart for Intensive Care - Electrocardiograms (MIMIC-IV-ECG) were evaluated. While the HiGHmed use case cardiology sensorik study captures fitness data including steps and heart rates (HR) through the Apple Watches warn by study participants, the MIMIC-IV-ECG dataset is an open source diagnostic records derived from the 12-leads electrocardiograms (ECG) of patients 10,11. The HiGHmed sensor data collection is structured such that the bulk of the continuous measurements are bundled as FHIR documents and transmitted through the health App to designated FHIR endpoints of the participating Medical Data Integration Centres (MeDICs). While some of the study sites retained the FHIR format, other sites bridged their endpoint to receive data in openEHR format. A recent study identified the lack of DQA tools to support FHIR bundles within the Medical Informatics Initiative (MII) 12. Also. previous studies have reported measurement inconsistencies and compliance issues as some of the quality concerns that must be addressed regarding the use of sensor data 13, 14. Therefore, this use case presents an opportunity to utilize the generalized DQA tool for the assessment of sensor data with FHIR and openEHR structures. Furthermore, the MIMIC-IV-ECG dataset has been widely used in building models for predicting various cardiac conditions such as myocardial infarction and rhythm abnormalities 15. However, it is important to determine the consistency of the cardiac annotations and the machine measurements that contribute to the accuracy of the model classifications 16. Considering that the MIMIC-IV-ECG dataset has distinct structure and rules from the HiGHmed sensor data, it allows us to assess the adaptability of the generalized DQA tool to new health data environments.\u003c/p\u003e \u003cp\u003e \u003cb\u003eObjective\u003c/b\u003e \u003c/p\u003e \u003cp\u003eIn this work, our goal is to develop a generalized DQA tool that is compatible with varied contradiction rules applicable to diverse health datasets, with validation focused on two distinct datasets from the cardiology domain. Specifically, the two structurally distinct datasets include: 1) the HiGHmed sensor data, and 2) MIMIC-IV-ECG dataset. The objective of this approach is to bring flexibility into DQA tool implementation such that it can be extended to diverse health data contexts (e.g. biobanking).\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Diverse Data Collections\u003c/h2\u003e \u003cdiv id=\"Sec4\" class=\"Section3\"\u003e \u003ch2\u003e2.1.1. HiGHmed Use Case Cardiology Sensorik Study\u003c/h2\u003e \u003cp\u003eThe HiGHmed use case cardiology sensorik study is a project within the MII in which novel health-IT solutions aimed at increasing access to patients\u0026rsquo; medical data for research are being developed ADDIN10. NThis study utilizes the Apple Watch to continuously record fitness metrics, specifically steps and heart rates (HR) 17,18. The integrated health application employs FHIR profiles as means of data exchange with the participating MeDICs. While two MeDICs (G\u0026ouml;ttingen and W\u0026uuml;rzburg) receive bundled resources as FHIR binaries, two other MeDICs (Cottbus and Hannover) bridged the FHIR endpoint to map the resources into openEHR compositions 19. The Kansas City Cardiomyopathy Questionnaire (KCCQ) was also documented to assess the health status of patients 20. Daily survey of body weight, resting HR and blood pressure (BP) were measured manually. At the G\u0026ouml;ttingen data collection site, six participants were onboarded while W\u0026uuml;rzburg enrolled about 50 participants. Cottbus and Hannover have two and nine study participants respectively. All MeDICs of the partner sites transferred either FHIR binaries or archetype query language (AQL)-resultset to G\u0026ouml;ttingen for a centralized data quality analysis. As depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the base64 encoded FHIR binaries in JSON format bundles a collection of FHIR documents containing the resources. The binaries were transformed and decomposed into separate FHIR documents using \u003cem\u003ePowerShell\u003c/em\u003e script. All relevant resources were extracted from the documents and loaded into R using \u003cem\u003eparse_fhir\u003c/em\u003e function linked to the DQ analysis tool. About 13,102 document bundles were extracted from W\u0026uuml;rzburg FHIR binaries with each document bundle storing up to 600 observations. This resulted in a total of 138,637 step measurements, 497,923 device HR measurements, 32 KCCQ responses, and 6,693 daily survey responses including manual HR, systolic and diastolic BP, and weight for 38 subjects in the analyzed dataset. Cottbus transmitted 509 AQL-resultset for 2 subjects and a total of 26,293 step, 113,063 HR observations, 11 KCCQ responses, and 840 survey were extracted using \u003cem\u003eparse_openehr\u003c/em\u003e function. In G\u0026ouml;ttingen, the decomposed FHIR resources from 1,553 documents included 152,391 device HR, 70,204 steps, 20 KCCQ responses, and 1,559 daily survey. Hannover transferred preprocessed openEHR composition with 181,148 heart rate measurements, 49,139 steps, 476 daily surveys, and 8 KCCQ responses. For our analysis, we considered three contradictory dependencies including: 1) conflicting device HR measurements, 2) inconsistencies in device and manual HR measurements, and 3) conflicting responses in KCCQ.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section3\"\u003e \u003ch2\u003e2.1.2. MIMIC- IV-ECG Dataset\u003c/h2\u003e \u003cp\u003eMIMIC-IV-ECG is an open source ECG diagnostic dataset with over 800,000 records of 160,000 patients documenting cardiology reports and associated machine-generated measurements 11. The waveform_note_links.csv and report.csv stored the metadata of the ECG records and cardiologist reports, such as study identifiers, subject identifiers, time of ECGs, note identifiers of cardiologist reports, International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) diagnosis codes, and nametags for the ECG files. In the machine_measurements.csv, a summary report of machine-generated evaluations, such as heart rhythm analysis and HR, was documented. While the dataset supports researchers in automating the evaluation of cardiac-related conditions 15, there are concerns about the consistency of the ICD annotations documented during diagnosis and the machine-generated measurements 16. For our evaluation in this work, we considered the ICD annotations of all cardiac electronic devices including implanted pacemakers and defibrillators captured in the MIMIC-IV-ECG dataset as well as the corresponding pacing rhythms in the ECG measurements estimated through the machine algorithm. The detailed ICD codes are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The consistency evaluations target two cases including: 1) ICD annotation of pacemakers and defibrillators (Z95.0, Z95.810, Z45.01, Z45.02, \u0026amp; Z45.09) that are linked to hospital stay (has_stay_id) with corresponding machine pacing rhythms, and 2) machine pacing rhythms without ICD codes.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDescription of International Classification of Diseases (ICD) annotation of pacemaker, defibrillator, and other cardiac devices.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eICD-10 Code\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDescription\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZ95.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003epresence of cardiac pacemaker\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZ95.810\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003epresence of automatic implantable cardiac defibrillator\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZ45.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eencounter for adjustment and management of cardiac pacemaker\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZ45.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eencounter for adjustment and management of automatic implantable cardiac defibrillator\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZ45.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eencounter for adjustment and management of other cardiac devices\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Design of Generalized DQA Framework\u003c/h2\u003e \u003cp\u003e \u003cp\u003eThe design of the generalized DQA (gDQA) framework depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e uses higher-order function in R programming that creates and returns another function as output 21. According to the documented requirements in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, the tool requires an overarching function to fulfill ER1 (elicited requirement) that accepts custom rules generated externally, and an inner function (ER2) that evaluates the rules on the target dataset. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, although the domain rules are plugged into the main tool with a call during execution, these rules are generated as an independent object using the rule generator function. This approach is intended to allow diverse health domains to define and integrate applicable contradiction rules suitable for the study data to be analyzed. This is a departure from current practice in existing tools, where rules and acceptable variables are fixed in the tools without the flexibility to define custom rules applicable to varied use cases. Also, apart from the user-defined rules, there are no additional required metadata to manually guide the operation of the tool on what rules to evaluate on selected items 2,5. In compliance with ER3 in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, all rules were created at the atomic level and labeled appropriately to make the identified anomalies traceable. The generated contradiction rules are evaluated on each row of the study data, with every violation cached and returned as output. Arguments in the rule generator include labels of the rule-set to be created and list of rule expressions defined within it. During execution, users select the respective labels while the DQA tool imports the rules listed within the selected label. The output of the rule generator serves as the inner function that evaluates the parsed study data to produce an assessment report. To enhance the usability of the tool as required in ER4, an interactive module is created in Shiny R. The module allows import of datasets of different formats including csv, openEHR, and FHIR document bundles as required in ER2. Users are guided with the Help function in defining desired custom rules according to ER1. Through the engagement of prospective users of the tool including data managers and study nurses, the requirements ER5 addresses users\u0026rsquo; preference for selection of specific custom rules from the created pool. This will streamline the assessment report to specific violations.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eRequirement elicitation (ER) for the generalized data quality framework.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRequirements\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDescription\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePriority\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eER1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ecustom rule generator\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRequired\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eER2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ediverse datasets import (csv, fhir bundle, openEHR composition)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRequired\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eER3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003edetailed and traceable output\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRequired\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eER4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eusability support (interactive module)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRequired\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eER5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003especific custom rule selection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDesired\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Contradiction Assessment in HiGHmed Sensor Data\u003c/h2\u003e \u003cp\u003eWe analyzed contradictions in the decomposed resources from FHIR binaries and similar items from openEHR compositions in the HiGHmed sensor data using the \u003cem\u003egDQA\u003c/em\u003e tool. The implemented rules addressed three contradictions including 1) conflicting measurements within heart rates (HR) measured using the device, 2) inconsistencies in device and manual HR measurements, and 3) conflicting responses in KCCQ. To evaluate the contradictions in HR measured using the device, the effective time of each measurement is considered. The time difference between the preceding and the successive measurements is derived. A time difference of zero seconds is considered implausible provided the affected HR measurements are not duplicated. This would result in contradictory measurements recorded at the same time-point, which could bias the average HR of the subject for the day. Also, for the contradictions between device-recorded HR and those measured manually, the last HR measured by the device at the nearest time to the time of the manually captured HR was compared. The third check examines the responses by the study participants to the KCCQ questionnaire items. An example of a conflicting response is when the subject chooses to be satisfied and unsatisfied with a question regarding the quality of life on the same response date. The overall implication of such contradictions is that they affect the derived total score of the subject's health status.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.4. Consistency Assessment in MIMIC-IV-ECG Dataset\u003c/h2\u003e \u003cp\u003eWe implemented two consistency rules in the MIMIC-IV-ECG dataset focusing on the correlations between ICD code annotations of pacemakers and defibrillators against the machine-detected pacing rhythms: 1) annotation-to-measurement consistency, and 2) measurement-to-annotation consistency. For rule 1), we defined Boolean consistency rule to check whether patients with ICD codes associated with cardiac electronic devices (specifically pacemakers and defibrillators) show corresponding evidence of pacing rhythm in their ECG measurements. The implemented Boolean rule 2) evaluates cases where the ECG measurements indicate pacing rhythm but lacked corresponding ICD documentation for implanted cardiac devices. The inclusion criteria in both cases is the presence of has_stay_id which indicates whether the ECG records are associated to patients\u0026rsquo; hospital stay. We estimated the total pacing records from the MIMIC-IV-ECG dataset as a sum of records that satisfy rule 1) and records that violates rule 2). The proportion of satisfactions or violations are determined by their frequency in the total pacing records.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Implementation\u003c/h2\u003e \u003cp\u003e All the elicited requirements (ER) in Table\u0026nbsp; \u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e with the \u003cem\u003erequired\u003c/em\u003e and \u003cem\u003edesired\u003c/em\u003e priority tags were implemented. The implementation source code can be found in the Gitlab repository of the project 22. The tool supports both predefined and custom rules as required by ER1. Users are not limited by the number of evaluated dependencies within items since the rule construct uses R extract operator (\u003cspan\u003e$\u003c/span\u003e) which can link all items within a name list. Users have the option to parse datasets in three data formats including csv, FHIR, and openEHR for analysis based on ER2. The \u003cem\u003eparse_fhir\u003c/em\u003e and \u003cem\u003eparse_openehr\u003c/em\u003e functions in the background extracts the resources within the FHIR document bundle and openEHR compositions and transform them into tabular form. A preview of the decomposed resources is enabled as shown in the interface in Fig.\u0026nbsp; \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e . ER3 was fulfilled by providing separate fields for users to enter labels for their custom rules and specify the rule conditions. A \u003cem\u003erule_processor\u003c/em\u003e function converts the user-defined rules into expressions that are executable in R. After running the analysis, the labels of each contradiction rule are used to describe violated rules in the assessment report. Figure\u0026nbsp; \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e shows the interface of the interactive module as required in ER4 with implemented features to support users without programming experience in fulfilling ER1, ER2, ER3, and ER5. \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Conflicting Device-Recorded Heart Rate Measurements\u003c/h2\u003e \u003cp\u003eThe analysis of simulteneously-recorded heart rates (HR) using the wearable device revealed measurement conflicts across all study sites. At the W\u0026uuml;rzburg site, 9,968 (2.1%) of 497,933 observations contained conflicting records as shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. The measurement intervals averaged 428 seconds, excluding outliers (0 seconds and 2,689,109). Zero-second epochs represented conflicting values in successive measurements while the maximum epoch resulted from prolonged device outage. The G\u0026ouml;ttingen site demonstrated fewer conflicts, with 250 (0.16%) conflicting observations among 152,391 HR measurements. The epoch distribution showed a minimum of 0 seconds, first quartile at 186 seconds, median of 289 seconds, mean of 310 seconds, third quartile at 359 seconds, and maximum of 69,121 seconds. This distribution indicates HR averaging windows predominantly fell between 186\u0026ndash;359 seconds, consistent with the time path visualization in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. At the Hannover site, 140 conflicting measurements were identified in 118,148 HR readings. The Cottbus site showed the lowest rate with only 54 (0.05%) conflicting observations among 113,063 measurements. These conflicting values represent measurement discrepancies rather than simple duplicates, as HR values at identical time-points differ. Although all study sites demonstrated high plausibility rates as shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, these conflicting measurements could potentially bias daily HR averages. While the Apple Watch system acknowledges potential sensor limitations that may produce anomalous readings, our findings confirm that such measurement conflicts were minimal across all study sites 17.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePlausibility rate for heart rate (HR) measurements\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy site\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTotal HR Observations\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePlausible Records\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eImplausible Records\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePlausible Rate (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eW\u0026uuml;rzburg\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e497923\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e487955\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e9968\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e97.99\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eG\u0026ouml;ttingen\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e152391\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e152141\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e250\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e99.84\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCottbus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e113063\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e113009\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e99.95\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHannover\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e181148\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e181008\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e140\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e99.92\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.3. Inconsistent Manual and Device-Recorded Heart Rates\u003c/h2\u003e \u003cp\u003eA total of 2,363 observations were collected across all study sites, representing paired manual and device-recorded HR from 47 study participants. As presented in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, analysis of these paired measurements showed varied patterns of agreement and discrepancy. The majority of the comparison revealed 1,186 device-recorded observations exceeding the manual measurements by less than 20 bpm. Conversely, in 853 observations, the manual measurements exceeded the device readings by 20 bpm. Perfect agreement between the manual and device measurements occurred in 182 observations. Notable discrepancies exceeding 20 bpm were observed in 142 observations, with the device measurements exceeding manual readings in 96 instances and manual measurements exceeding device readings in 46 instances. As illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, while most discrepancies observed are within 20 bpm, the presence of larger discrepancies in about 6% of the observations deserve attention. These substantial differences could potentially trigger unnecessary medical interventions if not properly verified with contextual evidence. Additionally such discrepancies visible to users through health applications may unnecessarily elevate anxiety regarding health status.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBreakdown of heart rate (HR) measurement difference between device and manual methods.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTotal HR Readings\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEqual Measurement\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDevice HR\u0026thinsp;\u0026gt;\u0026thinsp;20bpm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eManual HR\u0026thinsp;\u0026gt;\u0026thinsp;20bpm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDevice HR\u0026thinsp;\u0026lt;\u0026thinsp;20bpm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eManual HR\u0026thinsp;\u0026lt;\u0026thinsp;20bpm\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2363\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e182\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1186\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e853\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.4. Conflicting Responses to Cardiology Questionnaire\u003c/h2\u003e \u003cp\u003eFrom 47 subjects, conflicting responses were observed in two subjects (4.3%). Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e illustrates the pattern for one representative subject who provided contradictory responses on the same study day. The conflicts were observed in two distinct KCCQ items: 1) subject\u0026rsquo;s ability to run fast or hurry up, and 2) subject\u0026rsquo;s quality of life. The subject simultaneously on the same study day indicated being \u0026ldquo;hindered a little\u0026rdquo; and \u0026ldquo;moderately hindered\u0026rdquo; while hurrying up or running fast. For the question about satisfaction with living the remainder of life with heart failure, the same subject submitted both \u0026ldquo;satisfied\u0026rdquo; and \u0026ldquo;unsatisfied\u0026rdquo; responses on the same day. These conflicting responses compromise the accuracy of the derived overall health status score, which is calculated from the KCCQ responses. Such inconsistencies introduce uncertainty in the assessment of the subject\u0026rsquo;s true health status and could potentially lead to inadequate clinical decisions if not verified. According to the study protocol, subjects were required to complete the KCCQ at scheduled time-points: on the first day of enrollment, three times at four-week intervals, and subsequently at three-month intervals. The observed conflicts reveal limitations in the data collection, specifically the lack of controls in the health application to prevent or flag multiple submissions on a single response date.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Consistency of Cardiac Annotation and Machine-Detected Rhythm\u003c/h2\u003e \u003cp\u003eOur analysis revealed that only 26,990 records out of a total of 800,035 records in the MIMIC-IV-ECG dataset have pacing detection. As presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, among these 26,990 records with any pacing detection, 20,652 records show consistent match between ICD annotation and machine-detected pacing rhythm. The remaining 6,338 records were cases where there are machine-pacing rhythm without corresponding ICD codes for the cardiac devices. The implication of this annotation gap is that ICD codes alone are not sufficient to determine the presence of pacemakers or defibrillators. Model reliance on solely ICD codes as ground truth may lead to missing actual pacing cases.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003e \u003cem\u003ePrincipal Contributions\u003c/em\u003e \u003c/p\u003e \u003cp\u003eOur research introduces \u003cem\u003egDQA\u003c/em\u003e, a reusable DQA framework that represents a significant shift from the constraints of predefined rules prevalent in existing contradiction assessment tools. By designing a flexible rule parsing system, \u003cem\u003egDQA\u003c/em\u003e accepts custom contradiction rules while maintaining a consistent assessment. By evaluating \u003cem\u003egDQA\u003c/em\u003e with distinct health datasets (HiGHmed sensor data and MIMIC-IV-ECG) of different structures and contradiction rule contexts, we demonstrated the adaptability of the tool for diverse health datasets and rules. This validation is particularly relevant considering the challenge of maintaining a consistent DQA in diverse health data environments. Though contradiction rules in different health domains vary however, the assessment methods implemented through \u003cem\u003egDQA\u003c/em\u003e complement the established standardized indicators proposed by Kahn et al. and Schmidt et al. 6,7. \u003cem\u003egDQA\u003c/em\u003e extends the utility of the harmonized frameworks by providing a flexible implementation. With the decentralization of the assessment rules parsed into the tool, the generalized framework supports researchers to define and integrate domain-specific quality checks while maintaining an alignment with established DQ standards. The applicability of the tool for datasets of varied formats is an evidence that our approach supports multi-site collaborations. This is particularly relevant in an era where data sharing across health disciplines transcends institutional and national boundaries 23. Furthermore, in clinical trial settings, where timely quality assessment is essential for monitoring data integrity throughout the data collection process, \u003cem\u003egDQA\u003c/em\u003e offers particular advantage. The framework’s support for custom rules enables trial-specific quality checks to be implemented rapidly, facilitating near real-time assessment report 24.\u003c/p\u003e \u003cp\u003e \u003cem\u003eData Quality Issues in Medical Wearables\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe \u003cem\u003egDQA\u003c/em\u003e tool identified three contradiction patterns in the sensor datasets. The inconsistencies in HR measurements when using a device against the manual method can lead to confusion in treatment options or unnecessary intervention by the subjects to address sudden spike. While it is understandable that the device returns an estimated average of HR measurements over a scheduled interval, two different estimates at the same time-point from the same device are controversial and require further investigation. In addition, responses to predefined cardiology questionnaires serve a crucial purpose in determining subjects' health status at different activity levels. Therefore, providing an accurate response at the scheduled interval is important. As observed in the findings, the few reported gaps in the questionnaire were due to unrestricted data entry outside the scheduled response interval.\u003c/p\u003e \u003cp\u003e \u003cem\u003eInconsistency in ICD Annotation and ECG Machine Measurements\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe inconsistencies observed in about 23% of the pacing records is an opportunity to improve the validation of ICD annotations in real-world applications. However, these cases may also be linked to undocumented temporary pacing devices during the patients’ hospital stay or devices implanted at different hospital not captured in the evaluated medical records. An expert review of these findings would confirm if they are true pacing or false positives from the machine algorithm. For model development, it is encouraged to consider both ICD codes and machine measurements as complementary signals.\u003c/p\u003e \u003cp\u003e \u003cem\u003egDQA and Existing Tools\u003c/em\u003e \u003c/p\u003e \u003cp\u003eA common gap identified in existing DQA frameworks is that assessment rules are predefined within the tools with extra work required for metadata definition to guide the evaluation of already embedded rules, which is time-consuming 2,5. This has been addressed by reducing the attributes of the \u003cem\u003egDQA\u003c/em\u003e tool to user-prepared study data and custom DQA rules. Users are aided with the populated item list to select from when defining rules to minimize manual efforts. Our approach shows that rather than modifying DQA tools to incorporate new domain rules, only the rule list is expanded as desired. This study has demonstrated that similar to the harmonization of data quality indicators, the tool implementing the indicators can equally be generalized. The findings in this study revealed contradictions within isolated items which Cho et al. 13 classified as an atemporal plausibility of surrounding measurements. However, while the averaging time window influenced the contradictory HR findings, the conflicting questionnaire responses (also isolated) were illogical owing to multiple data entry on same response date. Therefore, when the event times of the conflicting findings are compared without any reference to the actual measurements or responses, they already qualify for duplicated entries, which is another DQ indicator. The actual measurements and responses were only required to validate the contradictions. Hence, these findings are simply contextual with the measurement time. This is an indication that as timestamps provide context to rule out contradictions within interdependent items, isolated or interdependent items can also be contextual to timestamps 25.\u003c/p\u003e \u003cp\u003e \u003cem\u003eLimitations\u003c/em\u003e \u003c/p\u003e \u003cp\u003eWhile efforts have been put into making the \u003cem\u003egDQA\u003c/em\u003e tool user friendly, data preparation involving patient records distributed across multiple tables requires varied levels of expertise among potential users. Additionally, to reduce the reliance on domain expertise for effective rule definition, the current implementation will benefit from a large collection of established assessment rules that cut across different health domains. This can only be achieved through collaborations with domain experts. In future work, we envisage that rule generation can be automated based on the characteristics of the items in datasets and domain ontologies to limit human efforts in rule processing. Further evaluations performed in diverse clinical settings (e.g. support for DICOM, genomic datasets) will strengthen the performance of the tool by identifying potential improvements.\u003c/p\u003e "},{"header":"Conclusion","content":"\u003cp\u003eThe development of a reusable DQA tool that is compatible with multiple health datasets is a significant contribution in health data management. The agnostic nature of the tool to distinct rules as tested on sensor data from medical wearables, and the MIMIC-IV-ECG dataset demonstrates its adaptability. Also, the freedom of definition of assessment rules empowers domain experts to define effective rules on ad-hoc basis while maintaining established DQ standards without being limited by predefined rules.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cem\u003eAcknowledgements:\u0026nbsp;\u003c/em\u003eThe work is jointly funded by MII, NUM, grant number 01KX2121 and DZHK, grant number 81Z0300108\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eConflict of Interest:\u0026nbsp;\u003c/em\u003eThe authors declare that there is no conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eEthical approval and consent:\u0026nbsp;\u003c/em\u003e Use of HiGHmed sensor data falls under the Medical Informatics Initiative (MII) data quality assurance covered by the ethics committee votes (09.10.2020) for the project.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eMari\u0026ntilde;o J, Kasbohm E, Struckmann S, Kapsner LA, Schmidt CO. R Packages for Data Quality Assessments and Data Monitoring: A Software Scoping Review with Recommendations for Future Developments. Applied Sciences. 2022 Apr 22;12(9):4238,DOI:10.3390/app12094238. \u003c/li\u003e\n\u003cli\u003eTute E, Scheffner I, Marschollek M. A method for interoperable knowledge-based data quality assessment. BMC Med Inform Decis Mak. 2021 Dec;21(1):93.\u003c/li\u003e\n\u003cli\u003eBlacketer C, Defalco FJ, Ryan PB, Rijnbeek PR. Increasing trust in real-world evidence through evaluation of observational data quality. J Am Med Inform Assoc. 2021 Sep 18;28(10):2251\u0026ndash;7.\u003c/li\u003e\n\u003cli\u003eYusuf K, Tahar K, Sax U, Hoffmann W, Krefting D. Assessment of the Consistency of Categorical Features Within the DZHK Biobanking Basic Set. In: R\u0026ouml;hrig R, Grabe N, Hoffmann VS, H\u0026uuml;bner U, K\u0026ouml;nig J, Sax U, et al., editors. Studies in Health Technology and Informatics [Internet]. IOS Press; 2022 [cited 2022 Sep 10]. Available from: https://ebooks.iospress.nl/doi/10.3233/SHTI220809. \u003c/li\u003e\n\u003cli\u003eYusuf KO, Miljukov O, Han\u0026szlig; S, Schoneberg A, Wiesenfeldt M, Stecher M, et al. Consistency as a Data Quality Measure for German Corona Consensus items mapped from National Pandemic Cohort Network data collections. Methods Inf Med. 2023 Jan 3;a-2006-1086.\u003c/li\u003e\n\u003cli\u003eJohnson SG, Pruinelli L, Hoff A, Kumar V, Simon GJ, Steinbach M, et al. A Framework for Visualizing Data Quality for Predictive Models and Clinical Quality Measures. AMIA Jt Summits Transl Sci Proc. 2019;2019:630\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eStruckmann S, Mari\u0026ntilde;o J, Kasbohm E, Salogni E, Schmidt CO. dataquieR 2: An updated R package for FAIR data quality assessments in observational studies and electronic health record data. Journal of Open Source Software. 2024 Jun 28;9(98):6581. \u003c/li\u003e\n\u003cli\u003eKahn MG, Callahan TJ, Barnard J, Bauck AE, Brown J, Davidson BN, et al. A Harmonized Data Quality Assessment Terminology and Framework for the Secondary Use of Electronic Health Record Data. EGEMS (Wash DC). 2016 Sep 11;4(1):1244. \u003c/li\u003e\n\u003cli\u003eSchmidt CO, Struckmann S, Enzenbach C, Reineke A, Stausberg J, Damerow S, et al. Facilitating harmonized data quality assessments. A data quality framework for observational health research data collections with software implementations in R. BMC Med Res Methodol. 2021 Dec;21(1):63. \u003c/li\u003e\n\u003cli\u003eHaarbrandt B, Schreiweis B, Rey S, Sax U, Scheithauer S, Rienhoff O, et al. HiGHmed - An Open Platform Approach to Enhance Care and Research across Institutional Boundaries. Methods Inf Med. 2018;57(S 01):e66\u0026ndash;81. \u003c/li\u003e\n\u003cli\u003eGow B, Pollard T, Nathanson LA, Johnson A, Moody B, Fernandes C, et al. MIMIC-IV-ECG: Diagnostic Electrocardiogram Matched Subset [Internet]. PhysioNet; [cited 2025 May 8]. Available from: https://physionet.org/content/mimic-iv-ecg/1.0/\u003c/li\u003e\n\u003cli\u003eDraeger C, Tute E, Schmidt CO, Waltemath D, Boeker M, Winter A, et al. Identifying Relevant FHIR Elements for Data Quality Assessment in the German Core Data Set. In: Caring is Sharing \u0026ndash; Exploiting the Value in Data for Health and Innovation [Internet]. IOS Press; 2023 [cited 2025 Apr 18]. p. 272\u0026ndash;6. Available from: https://ebooks.iospress.nl/doi/10.3233/SHTI230117\u003c/li\u003e\n\u003cli\u003eCho S, Weng C, Kahn MG, Natarajan K. Identifying Data Quality Dimensions for Person-Generated Wearable Device Data: Multi-Method Study. JMIR Mhealth Uhealth. 2021 Dec 23;9(12):e31618.\u003c/li\u003e\n\u003cli\u003ePurta R, Hachen D, Striegel A, Poellabauer C, Lizardo O, et al. Exploring Compliance: Observations from a Large Scale Fitbit Study. In: Proceedings of the 2nd International Workshop on Social Sensing [Internet]. Pittsburgh PA USA: ACM; 2017 [cited 2024 May 3]. p. 55\u0026ndash;60. Available from: https://dl.acm.org/doi/10.1145/3055601.3055608. \u003c/li\u003e\n\u003cli\u003eStrodthoff N, Lopez Alcaraz JM, Haverkamp W. Prospects for artificial intelligence-enhanced electrocardiogram as a unified screening tool for cardiac and non-cardiac conditions: an explorative study in emergency care. European Heart Journal - Digital Health. 2024 Jul 1;5(4):454\u0026ndash;60.\u003c/li\u003e\n\u003cli\u003eMcKeen K, Oliva L, Masood S, Toma A, Rubin B, Wang B. ECG-FM: An Open Electrocardiogram Foundation Model [Internet]. arXiv; 2024 [cited 2025 May 9]. Available from: http://arxiv.org/abs/2408.05178.\u003c/li\u003e\n\u003cli\u003eApple Support [Internet]. 2023 [cited 2024 Mar 12]. Monitor your heart rate with Apple Watch. Available from: https://support.apple.com/en-us/HT204666.\u003c/li\u003e\n\u003cli\u003eApple Support [Internet]. [cited 2024 May 3]. Track daily activity with Apple Watch. Available from: https://support.apple.com/guide/watch/track-daily-activity-with-apple-watch-apd3bf6d85a6/watchos.\u003c/li\u003e\n\u003cli\u003eFHIR-Bridge. https://github.com/NUM-Forschungsdatenplattform/num-fhir-bridge\u003c/li\u003e\n\u003cli\u003eSpertus JA, Jones PG, Sandhu AT, Arnold SV. Interpreting the Kansas City Cardiomyopathy Questionnaire in Clinical Trials and Clinical Care: JACC State-of-the-Art Review. Journal of the American College of Cardiology. 2020 Nov 17;76(20):2379\u0026ndash;90.\u003c/li\u003e\n\u003cli\u003eWickham H. Introduction | Advanced R [Internet]. [cited 2025 Apr 17]. Available from: https://adv-r.hadley.nz/fp.html. \u003c/li\u003e\n\u003cli\u003eSource code for \u003cem\u003egeneralizedDQA\u003c/em\u003e development: https://gitlab.gwdg.de/medinfpub/kvf-group/generalized-dq-tool\u003c/li\u003e\n\u003cli\u003eHealth Data Space (EHDS) [Internet]. [cited 2024 Nov 26]. Available from: https://www.european-health-data-space.com/. \u003c/li\u003e\n\u003cli\u003eHouston L, Probst Y, Yu P, Martin A. Exploring Data Quality Management within Clinical Trials. Appl Clin Inform. 2018 Jan;9(1):72\u0026ndash;81. \u003c/li\u003e\n\u003cli\u003eYusuf KO, Chaplinskaya-Sobol I, Schoneberg A, Hanss S, Valentin H, Lorenz-Depiereux B, et al. Impact of Clinical Study Implementation on Data Quality Assessments - Using Contradictions within Interdependent Health Data Items as a Pilot Indicator. Stud Health Technol Inform. 2023 Sep 12;307:152\u0026ndash;8. \u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"University Medical Centre Göttingen","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Data quality, health data, medical wearables, FHIR, openEHR ","lastPublishedDoi":"10.21203/rs.3.rs-6929289/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6929289/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cb\u003eBackground\u003c/b\u003e\u003c/p\u003e \u003cp\u003eData quality assessment (DQA) in health research is guided by rules that express several indicators such as completeness, conformance, and plausibility. However, contradiction rules vary with diverse health domains. For example, while in cardiology, it is established that diastolic blood pressure should not exceed corresponding systolic measurements, in the biobank, it is an anomaly for the blood sample to remain stateless. Despite efforts to harmonize data quality indicators, implementations of DQA are often limited by predefined rules and number of evaluated interdependent items which makes tool reusability difficult. A generalized DQA framework that allows definition of custom contradiction rules compatible with datasets from varied health domains is yet to be reported.\u003c/p\u003e\u003cp\u003e\u003cb\u003eObjective\u003c/b\u003e\u003c/p\u003e \u003cp\u003eThe aim of this work is to develop a generalized DQA tool that can assess the quality of diverse health datasets, with validation focused on two distinct datasets of different structures and rules from the cardiology domain. Specifically, the validation datasets include 1) HiGHmed use case sensor data, and 2) the Medical Information Mart for Intensive Care - Electrocardiograms (MIMIC-IV-ECG) dataset.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMethods\u003c/b\u003e\u003c/p\u003e \u003cp\u003eA DQA tool that decentralizes assessment rule generation was developed in R. Among the elicited requirements for the tool development are: 1) support for custom rule definition, 2) support for diverse datasets and formats, 3) detailed and traceable assessment report, and 4) interactive interface to aid usability of the tool. To test support for diverse data formats, the tool incorporates both Fast Healthcare Interoperability Resources (FHIR) and open Electronic Health Record (openEHR) parsers in R that decomposes the bundled resources within the FHIR documents and openEHR compositions in the HiGHmed sensorik study. The generalizability of the DQA tool was tested using custom rules that are specific to the HIGHmed sensor data and the MIMIC-IV-ECG dataset. For usability of the DQA tool, a Shiny R interface was implemented for an interactive DQ assessment with choice datasets and user-defined rules.\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults\u003c/b\u003e\u003c/p\u003e \u003cp\u003eAll elicited requirements including those desired by target users were fulfilled. The generalized framework has two segments including an overarching function that allows custom rule generation and an inner function that evaluates the defined rules on target dataset. Dataset from the HiGHmed use case cardiology were first evaluated and subsequently, the tool was successfully reused for the MIMIC-IV-ECG datasets. Also, the tool evaluates contradictions using custom rules defined by study personnel and imported study data in the interactive module.\u003c/p\u003e\u003cp\u003e\u003cb\u003eDiscussion\u003c/b\u003e\u003c/p\u003e \u003cp\u003eA reusable DQA tool is offered that analyzes health data sourced from different data models to produce comparable results across multiple study sites. The interactive DQA interface would assist study personnel in monitoring study data to address emerging data quality concerns. The usability of the DQA tool means that domain experts have the freedom to define and integrate rules tailored to their domain.\u003c/p\u003e","manuscriptTitle":"A generalized Data Quality Assessment Framework for Diverse Health Datasets with varied Contradiction Rules","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-23 08:09:16","doi":"10.21203/rs.3.rs-6929289/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"7092d267-eca4-402c-ac0d-ae5a05f0f0e0","owner":[],"postedDate":"June 23rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":50285258,"name":"Medical Informatics"}],"tags":[],"updatedAt":"2025-06-23T08:09:16+00:00","versionOfRecord":[],"versionCreatedAt":"2025-06-23 08:09:16","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6929289","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6929289","identity":"rs-6929289","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}