Work

Get out what you put in: optimising electronic medical record data

Year: 2023

Type: article

Abstract: As clinicians and researchers, it is our responsibility to use the data available to us to optimise the care of our patients. These data are vast and should, in theory, be able to provide granular information to make appropriate diagnostic and management decisions to provide the best care for patients. However, these data can take many forms in electronic medical records (EMR) that are also not designed to output the information in a coordinated manner. Accordingly, the measurement of data in EMR is often tenuous and slow. Now, with many institutions transitioning from handwritten to EMR software, foresight and a coordinated approach is required to transition to the digital age. Furthermore, EMR data has the potential to facilitate future audit and research activities through permitting the expediate collection of epidemiological data to predict outcomes and possibly select treatments.1 There is, however, a disconnect in the EMR design process, between software engineers and clinicians, which is undermining quality improvement initiatives. The design, implementation and operationalisation of functional systems that collects, exchanges and uses patient data to promote health care practice evolution and improved quality care has not been systematic and is poorly transparent.2 Medical text data are recorded electronically for multiple reasons, including as a means of communication, as a legal record, and for subsequent unit audit and research activities. Hospital audit activities are vitally important, and are necessary to maintain and improve the quality of hospital services.3 Research activities facilitate the development of novel strategies and treatments to improve care. Both of these activities are vital to hospital functioning and growth, and both require data. The type of data that can facilitate these activities is ideally standardized, clean and comprehensive. ICD10 codes are often used as a proxy by administrators for such data and whilst they are the best available option, they have significant flaws in their accuracy, particularly with respect to hospital acquired complications and are not granular enough to be used as a quality comparator.4 The stereotypical clinical workflow involves patient interaction, data input, integration and output/ extraction (both correspondence and audit quality indicators/performance metrics) and can either be challenged or optimised with EMR integration. A concerted approach to EMR operationalisation can optimise all aspects of this workflow by; modulating data input, and artificial intelligence (AI) incorporation to streamline data integration and output. Electronic medical record modification is a strategy that can be employed to optimise data input and increase the utility of routinely collected medical data. Through the use of online forms, mandatory fields and carefully designed proformas, it is theoretically possible to standardize the recording of medical text data. However, these strategies may not succeed in practice due to the workflows of staff members that are required to use these methods of documentation. For example, when online forms are developed in a manner that is cumbersome, staff may frequently develop means to complete their work in an efficient manner that circumvent the desired effect of the document structure. For example, when a given field is mandatory, a Hospital Medical Officer(HMO) under time pressure may put a single character in the field, so that it is marked as complete, while completing the majority of documentation in a single field of a given document. Additionally, making EMR modifications may be a slow and time-consuming process. Routinely collected data may also be optimised by incorporation of recordings/dictations for telehealth consults and multi-disciplinary meetings and ensuring compatible data sharing with all services from primary to quaternary level care. The use of templates is a comparatively simple means of improving data input and the value of routine medical data collection. By providing structured text for certain types of documentation (such as in the form of an 'acronym expansion' or a distributed document), healthcare workers are not only provided with a means by which their routine data collection may be standardized, but a recommendation regarding the components of an assessment that should take place. Offering a healthcare professional this template may also serve as a 'checklist' for given documents. For example, when a HMO is provided with a template for discharge summaries, this provides the HMO with guidance as to the type and level of information required in such discharge summaries. 'Checklist' approaches have previously been successful in other areas in surgery.5 Other advantages of template strategies include their ease of usability (as opposed to EMR modification) and agility (able to be modified at a departmental or institutional level quickly in response to changing clinical environments). In addition to discharge summaries, other examples of times when a template-based strategy may be useful include preadmission notes (see example Supplementary Information 1), ward round notes, and multi-disciplinary meeting notes. AI technologies including natural language processing (NLP) and generative AI are additional strategies that may be employed to improve data output, particularly when data input is unable to be optimised. NLP involves the application of computers to human language – either as speech or as text. Previous studies have identified that NLP can obtain meaningful unit audit activity data from medical free-text, such as discharge summaries.6 Furthermore, NLP has also been found to reduce the amount of missing crucial data (such as vital signs) within unstructured notes in electronic medical records.7 However, there are limitations to the level of performance that this type of method can obtain and, ultimately, the approach is limited to the information that is recorded, and what the users enter.8 In other words, if the data are not recorded, this method cannot obtain the information. Large language models (LLM) are another type of AI that has been successfully integrated into the clinical workflow to optimise output, specifically, producing patient clinic letters with high overall correctness, humanness and at a reading level that is broadly similar to current real-world human-generated letters.9 These strategies, either used in isolation or in combination, require a supportive local environment and vested stakeholders to be successful. Senior clinical staff have a critical role to play in the effective use of these strategies, ensuring hospital preparedness by liaising with appropriate senior administerial and information-technology representatives. The ongoing encouragement of junior staff to utilize the endorsed methods of data recording is also required. If the system is to work, there needs to be quality assurance (via clinically appropriate and regular audits) with senior clinician input, and not just set and forget at the start. Similarly, junior staff have a responsibility to support each other and should be incentivized to maintain the level of documentation that is required and engage in the audit process. Ultimately, an understanding that the standardized recording of information is an imperative role in the maintenance and optimisation of care should be fundamental to these initiatives. Medical text data are recorded every day in large quantities at institutions throughout the world. The transition from handwritten to EMR is inevitable, and clinicians need to embrace this process of evolution and engage in its implementation and improvement. Otherwise, it will be deferred to non-medical personnel at the potential cost of advancing patient care. Clinician engagement in EMR optimisation, using their inherent medical expertise and guided by evidence, can advocate for systems improvements in critical areas that might otherwise be forlorn by those without a clinical background including; transferability of data between medical systems (hospital networks and community practices), telehealth/conferencing integrations, clinical decision support system incorporation and adequate training modules for prevocational students.10, 11 Regarding this transition, local systems will dictate whether the current incremental change occurring currently is sufficient or if there should be an overhaul. This transition however must occur in parallel with appropriate cybersecurity implementation. Health institutions have a medico-legal and ethical responsibility to ensure the medical information adheres to good information security management practices to ensure confidentiality and integrity is assured. The information must also be processed and protected at a level commensurate with the classification to mitigate internal and external threats such as data leaks. Access to information must be regulated in accordance with legislative and business requirements, classifications and accessed only on a 'need to know', 'least privilege' principle and justifiable business means. Cryptographic controls are one example of how confidentiality and data may be protected, where cryptographic algorithm, protocols and key lengths are configured in accordance with Federal Government's Information Security Manual. Through the use of strategies including NLP, EMR optimisation, and templates, the utility of this routinely recorded medical data could be increased substantially. These data, obtained through routine activity, may provide novel insights to improve the outcomes of our patients. Open access publishing facilitated by The University of Adelaide, as part of the Wiley - The University of Adelaide agreement via the Council of Australian University Librarians. Brandon Stretton: Conceptualization; methodology; visualization; writing – original draft; writing – review and editing. Robert Padbury: Supervision; validation; visualization; writing – review and editing. Markus Trochsler: Supervision; validation; writing – review and editing. Thomas J. Hugh: Conceptualization; project administration; supervision; validation; visualization; writing – review and editing. Guy Maddern: Conceptualization; project administration; supervision; validation; visualization; writing – review and editing. Mark Boyd: Conceptualization; project administration; resources; supervision; validation; visualization; writing – review and editing. Lewis Hains: Visualization; writing – review and editing. Aashray Gupta: Methodology; supervision; visualization; writing – original draft; writing – review and editing. Joshua Kovoor: Conceptualization; investigation; visualization; writing – original draft; writing – review and editing. Stephen Bacchi: Conceptualization; supervision; validation; visualization; writing – original draft; writing – review and editing. Savio Barreto: Supervision; validation; visualization; writing – review and editing. Patrick G. O'Callaghan: Supervision; validation; visualization; writing – review and editing. Bianca Wong: Conceptualization; supervision; validation; visualization; writing – review and editing. Elizabeth Murphy: Supervision; validation; writing – review and editing. This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Supplementary Information 1 Example preadmissions template (Figure). Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.

Source: ANZ Journal of Surgery

DOI: 10.1111/ans.18559