The capabilities of natural language processing (NLP) methods have expanded significantly in recent years, and progress has been particularly driven by advances in data science and machine learning. However, NLP is still largely underused in patient-oriented clinical research and care (POCRC). A key reason behind this is that clinical NLP methods are typically developed, optimized, and evaluated with narrowly focused data sets and tasks (eg, those for the detection of specific symptoms in free texts). Such research and development (R&D) approaches may be described as problem oriented, and the developed systems perform specialized tasks well. As standalone systems, however, they generally do not comprehensively meet the needs of POCRC. Thus, there is often a gap between the capabilities of clinical NLP methods and the needs of patient-facing medical experts. We believe that to increase the practical use of biomedical NLP, future R&D efforts need to be broadened to a new research paradigm—one that explicitly incorporates characteristics that are crucial for POCRC. We present our viewpoint about 4 such interrelated characteristics that can increase NLP systems’ suitability for POCRC (3 that represent NLP system properties and 1 associated with the R&D process)—(1) interpretability (the ability to explain system decisions), (2) patient centeredness (the capability to characterize diverse patients), (3) customizability (the flexibility for adapting to distinct settings, problems, and cohorts), and (4) multitask evaluation (the validation of system performance based on multiple tasks involving heterogeneous data sets). By using the NLP task of clinical concept detection as an example, we detail these characteristics and discuss how they may result in the increased uptake of NLP systems for POCRC.JMIR Med Inform 2021;9(9):e18471
Health informatics is an emerging interdisciplinary field that has undergone considerable evolution over recent years. This evolution has largely been driven by the availability of big data and progress in artificial intelligence, machine learning, and data science . Big data from electronic health records (EHRs) have enabled researchers to train and execute neural network–based machine learning (eg, deep learning) algorithms for targeted problems, which have sometimes achieved performances that are comparable to those of human experts [ , ]. Clinical natural language processing (NLP)—one of the most complex subfields of health informatics—has also undergone rapid progress recently, which has been propelled by advanced machine learning, including deep learning [ ] and text representation methods [ , ]. Clinical NLP holds particular promise for improving evidence-based, patient-oriented clinical research and care (POCRC), since significant volumes of knowledge regarding patients and research evidence are encapsulated in the form of free text [ , ]. Patient-centered medicine and patient-oriented research focus on the unique needs and characteristics of patients in addition to the specialized skills of domain experts and the best available research evidence [ - ]. Due to its emphasis on outcomes that are important to patients, the POCRC model has been suggested to be superior in terms of quality compared to disease-oriented models, which focus on surrogate end points such as laboratory measurements and physical signs [ - ]. There has therefore been a continuous push, particularly in the practice of evidence-based medicine, to promote POCRC.
NLP tools and methods are traditionally optimized and evaluated based on their abilities to perform specialized, problem-specific, site-specific technical tasks. Such methods typically lack the capabilities to go beyond the problems that they are developed for and are unable to describe the relevant diverse characteristics of individual patients or help medical experts with patient-oriented decision-making. For example, studies on the fundamental NLP task of clinical concept detection (ie, concepts from EHRs or other sources) are typically designed to detect or extract small sets of disease-specific or problem-specific homogeneous concepts and are evaluated intrinsically via metrics such as accuracy and the F-measure. Such concepts, for example, include health conditions such as obesity , bleeding [ ], and drug reactions [ ] and behavioral patterns such as tobacco [ ] and alcohol [ ] use. Velupillai et al [ ] explained that although such systems may show high performances in intrinsic evaluation, they may have reduced value at the higher patient level. When the abovementioned problem-oriented NLP models are viewed through the lens of the well-defined model of patient-centered health care [ ], they appear to be analogous to disease-oriented, evidence-based medicine models, as they focus on a particular disease or problem instead of holistically taking patients into account. Such problem-oriented NLP research and development (R&D) has resulted in the creation of state-of-the-art models for many clinical text processing tasks and is essential for incorporating NLP progress into health informatics. However, NLP methods’ inability to meet the diverse requirements of medical experts has restricted their utility in POCRC. In a clinical scenario, particularly at the point of care, it is generally unrealistic to expect medical experts to customize and use multiple complex NLP methods to fully characterize patients based on the free-text information in patients’ EHRs. As a consequence of these limitations, the transition of clinical NLP systems from their R&D environments to regular use by medical experts has been slow and limited [ , ]. By building on recent advances, clinical NLP R&D has the potential to progress from the use of disease- and problem-oriented models to the use of patient-oriented models, provided that the needs from an NLP perspective are clearly defined. The gap between the capabilities of NLP systems and the POCRC needs of medical experts may be due to the lack of specification regarding what a patient-oriented perspective for clinical NLP should comprise and how patient-oriented clinical NLP systems can complement traditional problem-oriented systems. There have been little to no formal schemes, definitions, or discussions in medical informatics literature about the aspects of patient-orientedness for NLP. Given the explosive recent advances in NLP, it is now crucial to establish the building blocks of the requirements of patient-oriented NLP, so that methodological research may be targeted to directly improve POCRC. In the following paragraphs, we attempt to formulate what aspects should be considered when developing patient-oriented NLP systems.
Key NLP Needs for POCRC
Interpretability as a Core System Component (Interpretability)
Recent advances in machine learning, particularly in deep learning, have resulted in their successful application to specific clinical tasks [, ], and while most studies have relied on structured data from EHRs, some have used free-text information [ , , ]. Some studies have even generated patient representations based on the nonlinear transformations of all encoded information in EHRs [ ]. Despite the excellent results obtained by these systems in some cases, an obstacle to using these systems for POCRC—specifically when free text is involved—is the lack of interpretability. In fact, understanding how deep neural networks make their decisions is an area of active research in computer science [ , ]. Automation without interpretability means that the basis of a forecast or decision that is made by a system cannot be deciphered or explained by a medical expert. The inability to interpret the reasons behind automated systems’ decisions results in the inability of patient-facing medical experts to communicate these reasons to patients for tasks such as shared decision-making.
When designing and developing clinical NLP systems, informaticians must consider interpretability as a necessary constraint. Black-box models may be effective for a given task, but unless the decisions of a system are traceable in the desired manner, their application may not evolve beyond the problem-specific task for which they were developed . One method for potentially addressing this issue is integrating reporting mechanisms with machine learning models, so that the outputs of a task are not only predictions and numeric performance metrics but also modular reports that attempt to explain the reasons behind the predictions (eg, “which span of text in the note did the system think matched with concept X?” or “what were the top features that contributed to the system’s decision?”). The hypothetical framework depicted in illustrates the generation of reports by a system alongside other outputs, such as performance metrics. Such reporting mechanisms are uncommon in current clinical NLP systems, as the focus of R&D is almost invariably on some type of problem-specific performance metric. This is one aspect in which involving clinical stakeholders in the development process is essential, as clinical interpretability needs may be distinct from mathematical or statistical interpretability needs [ , ].
Broadening the Scopes of Clinical NLP Systems (Patient Centeredness)
We envision that clinical NLP systems will see greater adoption and use by medical experts for POCRC if their scopes are broader and are centered on patients rather than problems. For example, in the task of clinical concept detection, the ideal NLP systems for domain experts (and, hence, the patients they serve) would be those designed to detect ad hoc clinical concepts in free text (as specified by the expert) rather than a set of homogenous concepts. Using the current problem-oriented NLP systems perhaps adds to the burden imposed on experts, such as the burden of the “4000 clicks per shift”  problem, and contributes to burnout [ ]. In practice, patient-oriented researchers and caregivers require a holistic view of a patient, and from the perspective of clinical concept detection, such a representation of patients requires the detection of diverse information from patients’ EHRs. Such information may range from typical concepts that past NLP research has focused on, such as diseases or symptoms, to atypical concepts such as descriptions of daily life interactions that affect the mental and physical well-being of a patient. This is perhaps the key reason why structured EHR data are preferred and are commonly used for patient-level analytical and predictive tasks. Such data present a varied set of information that, when combined, provides a detailed representation of a patient [ ].
Future clinical NLP research that complements the existing advances in problem-based models should thus focus on developing frameworks that enable generalization at the patient level. For concept detection, this means enabling the specification of arbitrary clinical concepts of interest and detecting these concepts in the free-text portions of EHRs, which would result in the characterization of target patients based on these concepts. Since uncertainty is an inherent aspect of free text mining, instead of representing patient characteristics as binary variables, they can be represented by using continuous variables that represent the likelihood of a patient exhibiting specific characteristics (eg, the likelihood of viral exposure for a patient) . Such a framework for concept detection can, for example, facilitate the construction of research cohorts or be used to identify eligible subjects for study enrollment based on the diverse subject information that is encoded in free text. We have seen some recent research in clinical NLP naturally evolve to take this approach to concept detection and patient characterization. For example, Stubbs et al [ ] defined 13 variables, which involved diverse concepts that ranged from drug abuse to specified ranges of hemoglobin A1c levels, for identifying patients who meet the selection criteria for a clinical trial. Although this approach to patient characterization via NLP was not explicitly described by the authors as patient centered and contrasted with typical problem-focused approaches, it represents a natural evolution toward patient-oriented NLP systems because its parameter flexibility can be used for practical tasks. Ideally, the technical complexities of the NLP algorithms for concept detection (or other purposes) should be decoupled from the interface that medical experts use, so that they may focus on specifying their patient-oriented needs (eg, ad hoc clinical concepts) without having to learn how to use multiple systems or how to execute such algorithms in multiple environments. Building NLP systems that are generalizable in such a manner is not trivial by any means, but we believe that the time is now right for designing and developing clinical NLP frameworks that incorporate such broader scopes.
Flexible Systems Are More Likely to Stand the Test of Time (Customizability)
A problem that has been plaguing clinical NLP systems is the lack of customizability and adaptability. Many systems are so specialized to the problem-specific task for which they were designed that substantial effort is needed to adapt them to other tasks or data sources [, ]. The complexities of most clinical NLP systems, particularly those of recent systems that involve resource-heavy language models and intricate machine learning codes (eg, systems written in TensorFlow [ ]), are difficult for medical experts with non-NLP educational backgrounds to comprehend. As such, even for very similar tasks, such experts cannot customize previously developed systems to address the needs of new studies. We suspect that in most cases clinician researchers and caregivers do not even consider the possibility of diving deep into system source codes (eg, those of potentially customizable tools such as the Clinical Language Annotation, Modeling, and Processing Toolkit [ ]) and customizing them according to the specific needs of a study, as they are already burdened with information overload [ ].
Clinical NLP systems should thus focus on simplicity and customizability. Incorporating these aspects into the R&D of clinical NLP systems is also not trivial. However, they may be achieved by adhering to typical software development best practices. This may include using layered architectures, in which complexities are hidden under simple interfaces that expose users to customizable options. Such an architecture is shown in. In terms of clinical concept detection, the customizability of clinical NLP systems should enable medical experts to not only specify ad hoc concepts but also tune the system for different patient-oriented tasks (eg, cohort selection) by modifying system inputs, configurations, or parameters. Improving the customizability and simplicity of clinical NLP systems will undoubtedly increase their use in POCRC.
System Evaluations Using Multiple Data Sets With Heterogeneous Information (Multitask Evaluation)
System performance metrics obtained via evaluations based on a single data set can be misleading. Typical EHR-based free-text data sets are often constrained to small sets of patients with similar conditions, clinical settings, and social determinants, thereby causing systems that are built and evaluated based on such data sets to be overfit to the problem being studied . Furthermore, the unique characteristics of the site from which the EHRs originated, such as the focus of the entity (eg, an urban children’s hospital referral center) and the educational and training backgrounds of the note writers (eg, primary care physicians vs subspecialists), also influence how free text components are written. To gauge the true performances of clinical NLP methods, including performances associated with the three previously mentioned aspects, evaluations must be conducted based on multiple data sets with differing characteristic. The reuse utility of a system is substantially diminished if it is overfit to the characteristics of a specific data set. Reporting a system’s performance metrics (eg, the F-measure for concept detection) based solely on intrinsic evaluations of such specialized data sets may also be potentially perilous, since future users may incorrectly assume that the system will exhibit similar performances on other data sets. Consequently, the evaluation of systems based on multiple data sets with distinct characteristics is imperative for ensuring the robustness of systems.
To facilitate the greater adoption of NLP in POCRC, R&D models need to build on problem-oriented approaches and transition to patient-oriented ones. In this paper, we outlined the fundamental characteristics of patient-oriented NLP system design and development. We discussed 4 interrelated factors () that are essential—(1) interpretability, (2) patient centeredness, (3) customizability, and (4) multitask evaluation. We believe that given the rapid recent advances in data science, it is time to initiate a new paradigm for NLP R&D—one with a patient-oriented focus—in order to increase the impact that NLP R&D has on health care. Such a paradigm shift will require overcoming many barriers, which include, but are not limited to, challenges posed by informal texts, diversities in health-related languages [ ], the scarcity of annotated or labeled data, and difficulties that inhibit NLP systems’ progress from processing texts to understanding them [ ]. Recent advances in NLP, such as low-shot learning [ ], have the potential to aid researchers with the development of systems that are patient-oriented and, consequently, increase the impact of NLP in health care. This paradigm shift will be necessarily incremental, as researchers will build on and improve initial systems over time.
AS outlined the initial vision for this viewpoint paper. All coauthors contributed to the specification of the four factors discussed and contributed to the writing of this manuscript.
Conflicts of Interest
- Beam AL, Kohane IS. Big data and machine learning in health care. JAMA 2018 Apr 03;319(13):1317-1318. [CrossRef] [Medline]
- Larson DB, Chen MC, Lungren MP, Halabi SS, Stence NV, Langlotz CP. Performance of a deep-learning neural network model in assessing skeletal maturity on pediatric hand radiographs. Radiology 2018 Apr;287(1):313-322. [CrossRef] [Medline]
- Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA 2016 Dec 13;316(22):2402-2410. [CrossRef] [Medline]
- Shickel B, Tighe PJ, Bihorac A, Rashidi P. Deep EHR: A survey of recent advances in deep learning techniques for electronic health record (EHR) analysis. IEEE J Biomed Health Inform 2018 Sep;22(5):1589-1604 [FREE Full text] [CrossRef] [Medline]
- Mikolov T, Chen K, Corrado G, Dean J. Distributed representations of words and phrases and their compositionality. 2013 Dec Presented at: The 26th International Conference on Neural Information Processing Systems; December 5-10, 2013; Lake Tahoe, Nevada, USA p. 3111-3119.
- Devlin J, Chang M, Lee K, Toutanova K. BERT: Pre-training of deep bidirectional transformers for language understanding. 2019 Presented at: The 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies; June 2-7, 2019; Minneapolis, Minnesota p. 4171-4186. [CrossRef]
- Simmons M, Singhal A, Lu Z. Text mining for precision medicine: Bringing structure to EHRs and biomedical literature to understand genes and health. Adv Exp Med Biol 2016;939:139-166 [FREE Full text] [CrossRef] [Medline]
- Alsawas M, Alahdab F, Asi N, Li DC, Wang Z, Murad MH. Natural language processing: use in EBM and a guide for appraisal. Evid Based Med 2016 Aug;21(4):136-138. [CrossRef] [Medline]
- The advanced medical home: A patient-centered physician-guided model of healthcare. American College of Physicians. 2005. URL: https://www.acponline.org/acp_policy/policies/adv_medicalhome_patient_centered_model_healthcare_2006.pdf [accessed 2021-08-24]
- Evidence-Based Medicine Working Group. Evidence-based medicine. A new approach to teaching the practice of medicine. JAMA 1992 Nov 04;268(17):2420-2425. [CrossRef] [Medline]
- Greenhalgh T, Howick J, Maskrey N, Evidence Based Medicine Renaissance Group. Evidence based medicine: a movement in crisis? BMJ 2014 Jun 13;348:g3725 [FREE Full text] [CrossRef] [Medline]
- Vandermause R, Barg FK, Esmail L, Edmundson L, Girard S, Perfetti AR. Qualitative methods in patient-centered outcomes research. Qual Health Res 2017 Feb;27(3):434-442. [CrossRef] [Medline]
- Sacristán JA. Patient-centered medicine and patient-oriented research: improving health outcomes for individual patients. BMC Med Inform Decis Mak 2013 Jan 08;13:6 [FREE Full text] [CrossRef] [Medline]
- Godlee F. Outcomes that matter to patients. BMJ 2012 Jan 11;344(7839):e318. [CrossRef]
- Green AR, Carrillo JE, Betancourt JR. Why the disease-based model of medicine fails our patients. West J Med 2002 Mar;176(2):141-143 [FREE Full text] [Medline]
- Fleming TR, DeMets DL. Surrogate end points in clinical trials: are we being misled? Ann Intern Med 1996 Oct 01;125(7):605-613. [CrossRef] [Medline]
- de Grooth H, Parienti J, Oudemans-van Straaten HM. Should we rely on trials with disease- rather than patient-oriented endpoints? Intensive Care Med 2018 Apr;44(4):464-466 [FREE Full text] [CrossRef] [Medline]
- Lingren T, Thaker V, Brady C, Namjou B, Kennebeck S, Bickel J, et al. Developing an algorithm to detect early childhood obesity in two tertiary pediatric medical centers. Appl Clin Inform 2016 Jul 20;7(3):693-706 [FREE Full text] [CrossRef] [Medline]
- Li R, Hu B, Liu F, Liu W, Cunningham F, McManus DD, et al. Detection of bleeding events in electronic health record notes using convolutional neural network models enhanced with recurrent neural network autoencoders: Deep learning approach. JMIR Med Inform 2019 Feb 08;7(1):e10788 [FREE Full text] [CrossRef] [Medline]
- Li F, Yu H. An investigation of single-domain and multidomain medication and adverse drug event relation extraction from electronic health record notes using advanced deep learning models. J Am Med Inform Assoc 2019 Jul 01;26(7):646-654 [FREE Full text] [CrossRef] [Medline]
- Hegde H, Shimpi N, Glurich I, Acharya A. Tobacco use status from clinical notes using natural language processing and rule based algorithm. Technol Health Care 2018;26(3):445-456. [CrossRef] [Medline]
- Afshar M, Phillips A, Karnik N, Mueller J, To D, Gonzalez R, et al. Natural language processing and machine learning to identify alcohol misuse from the electronic health record in trauma patients: development and internal validation. J Am Med Inform Assoc 2019 Mar 01;26(3):254-261. [CrossRef] [Medline]
- Velupillai S, Suominen H, Liakata M, Roberts A, Shah AD, Morley K, et al. Using clinical natural language processing for health outcomes research: Overview and actionable suggestions for future advances. J Biomed Inform 2018 Dec;88:11-19 [FREE Full text] [CrossRef] [Medline]
- Carrell DS, Schoen RE, Leffler DA, Morris M, Rose S, Baer A, et al. Challenges in adapting existing clinical natural language processing systems to multiple, diverse health care settings. J Am Med Inform Assoc 2017 Sep 01;24(5):986-991 [FREE Full text] [CrossRef] [Medline]
- Assale M, Dui LG, Cina A, Seveso A, Cabitza F. The revival of the notes field: Leveraging the unstructured content in electronic health records. Front Med (Lausanne) 2019 Apr 17;6:66 [FREE Full text] [CrossRef] [Medline]
- Rajkomar A, Oren E, Chen K, Dai AM, Hajaj N, Hardt M, et al. Scalable and accurate deep learning with electronic health records. NPJ Digit Med 2018 May 08;1:18 [FREE Full text] [CrossRef] [Medline]
- Avati A, Jung K, Harman S, Downing L, Ng A, Shah NH. Improving palliative care with deep learning. BMC Med Inform Decis Mak 2018 Dec 12;18(Suppl 4):122 [FREE Full text] [CrossRef] [Medline]
- Xiao C, Choi E, Sun J. Opportunities and challenges in developing deep learning models using electronic health records data: a systematic review. J Am Med Inform Assoc 2018 Oct 01;25(10):1419-1428 [FREE Full text] [CrossRef] [Medline]
- Zhu R, Tu X, Huang J. Using deep learning based natural language processing techniques for clinical decision-making with EHRs. In: Dash S, Acharya B, Mittal M, Abraham A, Kelemen A, editors. Deep Learning Techniques for Biomedical and Health Informatics. Cham, Switzerland: Springer International Publishing; 2020:257-295.
- Miotto R, Li L, Kidd BA, Dudley JT. Deep patient: An unsupervised representation to predict the future of patients from the electronic health records. Sci Rep 2016 May 17;6:26094 [FREE Full text] [CrossRef] [Medline]
- Montavon G, Samek W, Müller K. Methods for interpreting and understanding deep neural networks. Digit Signal Process 2018 Feb;73:1-15 [FREE Full text] [CrossRef]
- Ghorbani A, Abid A, Zou J. Interpretation of neural networks is fragile. 2019 Jul 17 Presented at: The 33rd AAAI Conference on Artificial Intelligence; January 27 to February 1, 2019; Honolulu, Hawaii, USA p. 3681-3688. [CrossRef]
- Ching T, Himmelstein DS, Beaulieu-Jones BK, Kalinin AA, Do BT, Way GP, et al. Opportunities and obstacles for deep learning in biology and medicine. J R Soc Interface 2018 Apr;15(141):20170387 [FREE Full text] [CrossRef] [Medline]
- Hohman FM, Kahng M, Pienta R, Chau DH. Visual analytics in deep learning: An interrogative survey for the next frontiers. IEEE Trans Vis Comput Graph 2018 Jun 04. [CrossRef] [Medline]
- Hill RG, Sears LM, Melanson SW. 4000 clicks: a productivity analysis of electronic medical records in a community hospital ED. Am J Emerg Med 2013 Nov;31(11):1591-1594. [CrossRef] [Medline]
- Verghese A, Shah NH, Harrington RA. What this computer needs is a physician: Humanism and artificial intelligence. JAMA 2018 Jan 02;319(1):19-20. [CrossRef] [Medline]
- Goldstein BA, Navar AM, Pencina MJ, Ioannidis JPA. Opportunities and challenges in developing risk prediction models with electronic health records data: a systematic review. J Am Med Inform Assoc 2017 Jan;24(1):198-208 [FREE Full text] [CrossRef] [Medline]
- Sarker A, Klein AZ, Mee J, Harik P, Gonzalez-Hernandez G. An interpretable natural language processing system for written medical examination assessment. J Biomed Inform 2019 Oct;98:103268. [CrossRef] [Medline]
- Stubbs A, Filannino M, Soysal E, Henry S, Uzuner Ö. Cohort selection for clinical trials: n2c2 2018 shared task track 1. J Am Med Inform Assoc 2019 Nov 01;26(11):1163-1171 [FREE Full text] [CrossRef] [Medline]
- Johnson SB, Adekkanattu P, Campion TR, Flory J, Pathak J, Patterson OV, et al. From sour grapes to low-hanging fruit: A case study demonstrating a practical strategy for natural language processing portability. AMIA Jt Summits Transl Sci Proc 2018 May 18;2017:104-112 [FREE Full text] [Medline]
- Abadi M, Agarwal A, Barham P, Brevdo E, Chen Z, Citro C, et al. TensorFlow: A system for large-scale machine learning. 2016 Presented at: The 12th USENIX Conference on Operating Systems Design and Implementation; November 2-4, 2016; Savannah, Georgia, USA p. 265-283 URL: https://www.usenix.org/system/files/conference/osdi16/osdi16-abadi.pdf [CrossRef]
- Soysal E, Wang J, Jiang M, Wu Y, Pakhomov S, Liu H, et al. CLAMP - a toolkit for efficiently building customized clinical natural language processing pipelines. J Am Med Inform Assoc 2018 Mar 01;25(3):331-336 [FREE Full text] [CrossRef] [Medline]
- Klerings I, Weinhandl AS, Thaler KJ. Information overload in healthcare: too much of a good thing? Z Evid Fortbild Qual Gesundhwes 2015;109(4-5):285-290. [CrossRef] [Medline]
- Oleynik M, Kugic A, Kasáč Z, Kreuzthaler M. Evaluating shallow and deep learning strategies for the 2018 n2c2 shared task on clinical text classification. J Am Med Inform Assoc 2019 Nov 01;26(11):1247-1254 [FREE Full text] [CrossRef] [Medline]
- Sheikhalishahi S, Miotto R, Dudley JT, Lavelli A, Rinaldi F, Osmani V. Natural language processing of clinical notes on chronic diseases: Systematic review. JMIR Med Inform 2019 Apr 27;7(2):e12239 [FREE Full text] [CrossRef] [Medline]
- Xia C, Zhang C, Zhang J, Liang T, Peng H, Yu PS. Low-shot learning in natural language processing. 2020 Presented at: 2020 IEEE Second International Conference on Cognitive Machine Intelligence (CogMI); October 28-31, 2020; Atlanta, Georgia, USA p. 185-189. [CrossRef]
|EHR: electronic health record|
|NLP: natural language processing|
|POCRC: patient-oriented clinical research and care|
|R&D: research and development|
Edited by R Kukafka, G Eysenbach; submitted 28.02.20; peer-reviewed by L Li, N Karnik, MG Kim, V Foufi; comments to author 17.09.20; revised version received 13.02.21; accepted 02.08.21; published 28.09.21Copyright
©Abeed Sarker, Mohammed Ali Al-Garadi, Yuan-Chi Yang, Jinho Choi, Arshed A Quyyumi, Greg S Martin. Originally published in JMIR Medical Informatics (https://medinform.jmir.org), 28.09.2021.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Medical Informatics, is properly cited. The complete bibliographic information, a link to the original publication on https://medinform.jmir.org/, as well as this copyright and license information must be included.