This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), 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 http://medinform.jmir.org/, as well as this copyright and license information must be included.
The process of documentation in electronic health records (EHRs) is known to be time consuming, inefficient, and cumbersome. The use of dictation coupled with manual transcription has become an increasingly common practice. In recent years, natural language processing (NLP)–enabled data capture has become a viable alternative for data entry. It enables the clinician to maintain control of the process and potentially reduce the documentation burden. The question remains how this NLP-enabled workflow will impact EHR usability and whether it can meet the structured data and other EHR requirements while enhancing the user’s experience.
The objective of this study is evaluate the comparative effectiveness of an NLP-enabled data capture method using dictation and data extraction from transcribed documents (NLP Entry) in terms of documentation time, documentation quality, and usability versus standard EHR keyboard-and-mouse data entry.
This formative study investigated the results of using 4 combinations of NLP Entry and Standard Entry methods (“protocols”) of EHR data capture. We compared a novel dictation-based protocol using MediSapien NLP (NLP-NLP) for structured data capture against a standard structured data capture protocol (Standard-Standard) as well as 2 novel hybrid protocols (NLP-Standard and Standard-NLP). The 31 participants included neurologists, cardiologists, and nephrologists. Participants generated 4 consultation or admission notes using 4 documentation protocols. We recorded the time on task, documentation quality (using the Physician Documentation Quality Instrument, PDQI-9), and usability of the documentation processes.
A total of 118 notes were documented across the 3 subject areas. The NLP-NLP protocol required a median of 5.2 minutes per cardiology note, 7.3 minutes per nephrology note, and 8.5 minutes per neurology note compared with 16.9, 20.7, and 21.2 minutes, respectively, using the Standard-Standard protocol and 13.8, 21.3, and 18.7 minutes using the Standard-NLP protocol (1 of 2 hybrid methods). Using 8 out of 9 characteristics measured by the PDQI-9 instrument, the NLP-NLP protocol received a median quality score sum of 24.5; the Standard-Standard protocol received a median sum of 29; and the Standard-NLP protocol received a median sum of 29.5. The mean total score of the usability measure was 36.7 when the participants used the NLP-NLP protocol compared with 30.3 when they used the Standard-Standard protocol.
In this study, the feasibility of an approach to EHR data capture involving the application of NLP to transcribed dictation was demonstrated. This novel dictation-based approach has the potential to reduce the time required for documentation and improve usability while maintaining documentation quality. Future research will evaluate the NLP-based EHR data capture approach in a clinical setting. It is reasonable to assert that EHRs will increasingly use NLP-enabled data entry tools such as MediSapien NLP because they hold promise for enhancing the documentation process and end-user experience.
Electronic health records (EHRs) permeate most medical practices in the United States [
Since their inception, EHRs “have been proposed as a means for improving availability, legibility, and completeness of patient information” [
As EHR implementations continue, physicians frequently express dissatisfaction with EHR documentation methods and usability [
Natural language processing (NLP) has emerged as a viable solution for clinical data capture. Many challenges remain for keyboard-and-mouse entry, namely, having to type text and negotiate the often unwieldy EHR interface to record information in structured fields. This is exacerbated by the fact that much of the EHR content continues to be unstructured [
Although most clinical information in EHRs is stored as unstructured data, such as clinical narrative, its electronic capture or retrieval has been challenging [
Problems associated with the time required for documentation and usability are well established. However, there is also evidence to suggest that quality of EHR documents (eg, progress reports) is problematic [
In this study, we were focally concerned with testing NLP-enabled dictation-based data capture as a potential solution for relieving the increased burden of documentation. The benchmarks of performance include measures of time, data quality, and usability. According to Cimino [
A variety of modalities have been used for creating clinical documentation for EHR data capture or extraction to generate structured, actionable data (ie, data that are consumable, usable, reusable by a computer, and exchangeable with other computer systems in an efficient manner). These modalities include paper-based records transfer; verbal communication; direct entry or direct entry with macros; electronic templates; “Smart Forms”; dictation using speech recognition, sometimes known as voice recognition or continuous speech recognition; transcription or transcription with manual error correction; patient-recorded data (various methods); and hybrids, with or without NLP data capture, also termed “text processing” [
NLP encompasses a family of methods for processing text. These methods have been used for a range of EHR applications [
Whereas relatively few NLP systems for structured clinical data capture are implemented outside academic medical centers [
The NLP engine used by the MediSapien NLP data capture application is the Medical Language Extraction and Encoding System (MedLEE), which was developed at Columbia University in the Department of Biomedical Informatics. MedLEE accepts unstructured clinical text inputs and outputs structured clinical information in a variety of formats [
Developed by ZyDoc Medical Transcription, the MediSapien NLP data capture application allows doctors to use unstructured dictation to capture structured data in the EHR. MediSapien NLP preprocesses documents, leverages the MedLEE NLP engine, and postprocesses the NLP output using patent-pending processes that augment the NLP engine’s output. It also enables a workflow by which (1) the physician dictates, (2) the dictation is transcribed or subjected to speech recognition, (3) MediSapien NLP generates structured data from the transcription, and (4) the structured data and text are inserted into the EHR, although we simulated the EHR interface in the study.
It should be noted that this was a formative study designed to investigate the comparative effects of data capture methods enabled by the NLP system. The focus of the analysis was on characterizing interactive behavior and system usability rather than the NLP method. Future studies will investigate the efficacy of the NLP processes used by the system.
The objectives of this study were to (1) measure the effects, relative to using Standard Entry only, of using 3 NLP-based documentation protocols on EHR documentation time and quality and (2) measure the effects of an NLP Entry–based protocol and a Standard Entry–based protocol on the usability of the documentation process.
Five dictation-based electronic health record (EHR) data capture methods.
MediSapien NLP application user interface, illustrating the volume of structured data generated by MediSapien NLP. NLP: natural language processing.
The study evaluated an NLP-enabled solution for documentation. Specifically, we focused on three problem areas related to EHR data capture: (1) efficiency, including time required for data capture; (2) effectiveness, encompassing documentation quality; and (3) physician satisfaction, based on usability. We compared a novel dictation-based protocol using MediSapien NLP for structured data capture (“NLP-NLP”) against a standard, keyboard-and-mouse structured data capture protocol where the study participant was instructed to generate EHR documentation as in normal clinical practice (“Standard-Standard”) as well as 2 novel hybrid protocols (“NLP-Standard” and “Standard-NLP”) to determine which protocols provided better results in terms of data capture time, documentation quality, and physician satisfaction. The hybrid protocols were included because we anticipated that mixed forms and modalities of interaction may serve as realistic alternatives to a one-dimensional NLP approach or standard data entry. For example, certain parts of clinical notes may be better served by one modality of entry or the other; a note’s assessment and plan sections are often more given to free text and may therefore be suited to a dictation-based modality, whereas a note’s history and physical examination sections are less so and therefore may be better suited to a different modality. A hybrid approach may offer greater flexibility and can be adapted to the preferences of individual users. The study presented here is formative work that focused more directly on the nature of user interaction and the user experience rather than the efficacy or precision of the NLP system or the system for insertion of data in the EHR. These will be addressed in the next phase of research.
This study contrasted 4 conditions involving combinations of NLP Entry and Standard Entry (referred to in this paper as documentation protocols) on the following measures: documentation time, documentation quality, and usability of the documentation process.
The Standard Entry method (ie, how physicians typically use an EHR to document) entailed using a keyboard and mouse for typing text and negotiating the graphical user interface (eg, drop-down menus, check-boxes) to record information in structured fields.
In the NLP Entry method, the participants dictated the content of the documents. They did not enter any documentation using the keyboard or mouse. Their dictation was transcribed, and the transcription was inputted into the MediSapien NLP application. That application outputted a document containing structured data (an example of which is shown in
Documentation methods used for each documentation protocol.
Documentation protocols | Documentation method for history and physical examination | Documentation method for assessment and plan |
NLPa-NLP | NLP Entry | NLP Entry |
NLP-Standard | NLP Entry | Standard Entry |
Standard-NLP | Standard Entry | NLP Entry |
Standard-Standard (control) | Standard Entry | Standard Entry |
aNLP: natural language processing.
In the study, each physician was asked to document 4 notes using 4 methods including 1 control method (Standard-Standard protocol) and 3 experimental protocols consisting of combinations of NLP Entry and Standard Entry for documenting different parts of the note, as presented in
Physician participants were recruited through referrals. Two of the coauthors (BS and PS) referred us to several physicians who in turn made additional referrals. The inclusion criteria for the participants were as follows: (1) must be a neurologist, cardiologist, or nephrologist, the 3 specialties included in the study; (2) must be a senior resident, fellow, or attending; and (3) must be a current user of the Columbia University Medical Center’s (CUMC) Crown Allscripts EHR (Chicago, IL). The participants were each compensated US $500 for their efforts.
This study was conducted at CUMC. The test protocol was administered with physician participants at their offices. Fictitious patients were created for the study, and the participants documented their cases in a test environment of the Crown Allscripts EHR, which was the same EHR in which the participants documented during normal clinical practice. Participants were all experienced users of the system. The Crown Allscripts EHR had been in use in excess of 5 years at CUMC as of the time of the study. This test environment contained the same custom templates that participants used during normal clinical practice. As a result, the Standard Entry method simulated documentation during normal clinical practice as closely as reasonably possible.
The test scripts were based on anonymized transcription documents that were modified by 4 expert clinicians (2 fellows and 2 attending physicians). These clinicians were not participants in the study. The test scripts consisted of history and physical examination sections but excluded assessment and plan sections. After reviewing test scripts that described cases of the fictitious patients, the participants generated 4 multisectional consultation or admission notes using 4 documentation protocols (
First, each participant read the instructions for generating consultation or admission notes based on the 4 provided test scripts, an example of which is shown in
Second, the participant was asked to review the test scripts and to generate 4 notes from 4 test scripts, 4 examples of which are shown in
Third, for documentation in which NLP Entry was used, the participant’s dictation was transcribed; the transcription was processed by MediSapien NLP; and the transcription, structured data generated, and any documentation generated for the note by Standard Entry (if applicable) were combined to create the final note. A simulated interface and simulated note were used for NLP Entry: following a protocol, study assistants copied the generated unstructured and structured data into a Microsoft Word document to generate the final note. For Standard Entry, an actual EHR interface was used.
Finally, after reviewing their final notes, the participants completed 2 System Usability Scale (SUS) surveys [
Summary of usability scores (mean, SD) and paired
Usability question | Standard-Standard, mean (SD) | NLPa-NLP, mean (SD) | |
I think that I would like to use this method frequently for admitting notes. | 2.9 (0.9) | 3.3 (0.8) | .21 |
I found this method unnecessarily complex. | 2.5 (1.4) | 3.8 (0.8) | .003 |
I thought this method was easy to use. | 2.8 (1) | 4.2 (0.6) | <.001 |
I think that I would need assistance to be able to use this method. | 3.3 (1.1) | 3.6 (0.9) | .24 |
I found the various functions in the processes of the method were well integrated. | 2.6 (0.9) | 3.2 (1) | .05 |
I would imagine that most people would learn to use this method very quickly. | 3.0 (0.9) | 3.8 (0.7) | .01 |
I found this method very cumbersome/awkward to use. | 2.6 (1.1) | 3.7 (0.9) | .004 |
I felt very confident using this method. | 3.6 (0.8) | 3.4 (0.8) | .43 |
I would need to learn a lot of things before I could get going with this method. | 3.6 (1) | 3.8 (0.8) | .40 |
I feel the method would fit well in my existing workflow. | 2.8 (0.9) | 3.4 (0.9) | .08 |
aNLP: natural language processing.
Example of a neurology test script.
Example of part of the history and physical examination section of a neurology consultation note generated using the Standard-NLP protocol, illustrating the part of the note that was generated by Standard Entry. NLP: natural language processing.
Example of part of the history and physical examination section of a neurology consultation note generated using the NLP-Standard protocol, illustrating the part of the note that was generated by NLP Entry. NLP: natural language processing.
Example of part of the assessment and plan section of a neurology consultation note generated using the Standard-NLP protocol, illustrating the part of the note that was generated by NLP Entry. NLP: natural language processing.
Example of part of the assessment and plan section of a neurology consultation note generated using the NLP-Standard protocol, illustrating the part of the note that was generated by Standard Entry. NLP: natural language processing.
The time required for documentation was measured using a stopwatch.
The 4 expert clinicians (2 fellows and 2 attending physicians) were not participants and were blind to the protocols used to generate the documentation they evaluated. They were provided with gold standard versions of the test documentation they were asked to evaluate and told that the gold standard versions represented “high quality notes.” They were then instructed to measure documentation quality by comparing participants’ final test documentation against the gold standard versions of that documentation using the PDQI-9 tool [
The gold standard versions of the documentation were generated by the expert clinicians in Microsoft Word. They produced these documents from clinical notes and modified them so that they were consistent with the clinical profile of the patient (ie, the patient’s assessment and treatment were consistent and derivable from history and physical examination findings). The expert clinicians were instructed to ensure that all elements of the documents were internally consistent and that they truly reflected a gold standard. The expert clinicians were compensated at a rate of US $125 per hour for their efforts. We did not have access to the interim work product of the expert clinicians. We were only provided with the expert clinicians’ grades.
The usability of the documentation processes was assessed using a modified version of the SUS [
Physician Documentation Quality Instrument (PDQI-9) tool.
Data were analyzed using Intercooled Stata version 9.2 (StataCorp LP). Demographics were tabulated in regard to participants’ years of EHR experience, years of experience with dictation, the number of cases per subject area, the frequency of use of each of the 4 protocols, dictation time, usability scores, and quality scores. The Shapiro-Wilk W test was used to determine whether continuous variables were normally distributed. Results are presented as mean (SD) or analysis of variance (ANOVA) results, median (interquartile range), or percentage; respectively, comparisons were made using
Pearson correlation was performed on continuous variables. The association of years of EHR experience, years of experience with dictation, the 4 protocols (Standard-Standard, Standard-NLP, NLP-Standard, and NLP-NLP), and the 3 subject areas (cardiology, nephrology, and neurology) with the note dictation time was assessed using ANOVA.
Statistical significance was defined as alpha=.05 and Bonferroni correction was used where applicable for multiple comparisons.
The study was approved by the Institutional Review Board at Columbia University (#AAAK2458). All participants gave consent before their participation and were fully briefed on the true objectives of the study. The study protocol adhered to strict standards of confidentiality and privacy.
A total of 31 unique individuals documented 3.8 (SD 0.7) notes on average. Of these, 28 participants completed all 4 protocols, 2 participants completed 2 protocols, and 1 participant completed 1 protocol. The participants who did not complete all 4 protocols were called away from the study and therefore could not finish the task. These individuals had an average EHR experience of 6.6 (SD 3.4) years (data were available for 30 individuals) and an average dictation experience of 2.8 (SD 5.6) years (data were available for 29 individuals). There was a significant association between years of EHR experience and years of dictation experience (
A total of 118 notes were documented across the 3 subject areas of cardiology (22/118, 18.6%), nephrology (21/118, 17.8%), and neurology (75/118, 63.6%). The Standard-Standard, Standard-NLP, NLP-Standard, and NLP-NLP protocols were used in 28/118 (23.7%), 28/118 (23.7%), 30/118 (25.4%), and 32/118 (27.1%) documented notes, respectively. The frequency of use of the 4 protocols was balanced across the 3 subject areas (see
Frequency of use of the 4 protocols by subject area for each documented note.
Protocol | Documented cardiology notes, n (%) | Documented nephrology notes, n (%) | Documented neurology notes, n (%) | Total number of documented notes, n (%) |
Standard-Standard | 5 (23) | 5 (24) | 18 (24) | 28 (23.7) |
Standard-NLPa | 5 (23) | 4 (19) | 19 (25) | 28 (23.7) |
NLP-Standard | 6 (27) | 5 (24) | 19 (25) | 30 (25.4) |
NLP-NLP | 6 (27) | 7 (33) | 19 (25) | 32 (27.1) |
Total | 22 | 21 | 75 | 118 |
aNLP: natural language processing.
Median documentation time in minutes, with interquartile ranges, by protocol and subject area.
Protocol | Median (IQRa) time to document cardiology note (minutes) | Median (IQR) time to document nephrology note (minutes) | Median (IQR) time to document neurology note (minutes) |
Standard-Standard | 16.9 (16.5-19.7) | 20.7 (18.6-23.2) | 21.2 (17.6-29.9) |
Standard-NLPb | 13.8 (13.0-17.2) | 21.3 (14.5-29.8) | 18.7 (16.0-22.9) |
NLP-Standard | 7.5 (7.1-9.1) | 12.1 (10.7-12.2) | 11.0 (8.5-14.6) |
NLP-NLP | 5.2 (4.7-8.0) | 7.3 (6.6-9.1) | 8.5 (6.4-11.4) |
aIQR: interquartile range.
bNLP: natural language processing.
Interprotocol comparisons (Wilcoxon rank sum analysis).
Interprotocol comparisons | Statistical analysis of time difference ( |
||
Cardiology notes | Nephrology notes | Neurology notes | |
Standard-Standard vs Standard-NLPb | .60 | .81 | .20 |
Standard-Standard vs NLP-Standard | .01 | .03 | <.001 |
Standard-Standard vs NLP-NLP | .006 | .005 | <.001 |
Standard-NLP vs NLP-Standard | .006 | .05 | .001 |
Standard-NLP vs NLP-NLP | .006 | .008 | <.001 |
NLP-Standard vs NLP-NLP | .11 | .02 | .02 |
aStatistical significance level: alpha=.0083 after Bonferroni correction.
bNLP: natural language processing.
Document quality for each protocol (median values are presented).
Protocols and statistical comparisons | Document quality metricsa | |||||||||
A | T | U | O | C | S | Sy | I | Sum | ||
Standard-Standard (n=24) | 3.5 | 3 | 4 | 4 | 4 | 4 | 4 | 4 | 29 | |
Standard-NLPb (n=24) | 4 | 4 | 4 | 3.5 | 4 | 2.5 | 4 | 4 | 29.5 | |
NLP-Standard (n=27) | 4 | 3 | 3 | 3 | 3 | 3 | 4 | 4 | 26 | |
NLP-NLP (n=30) | 4 | 4 | 3 | 3 | 3 | 2 | 3 | 4 | 24.5 | |
Standard-Standard vs Standard-NLP | .04 | .03 | <.001 | |||||||
Standard-Standard vs NLP-Standard | .04 | .006 | ||||||||
Standard-Standard vs NLP-NLP | .002 | .02 | <.001 | .03 | ||||||
Standard-NLP vs NLP-Standard | .005 | |||||||||
Standard-NLP vs NLP-NLP | .02 | |||||||||
NLP-Standard vs NLP-NLP | .03 | .001 |
aThe 8 document quality metrics are as follows: Accurate, Thorough, Useful, Organized, Comprehensible, Succinct, Synthesized, and Internally Consistent.
bNLP: natural language processing.
cStatistical significance level: alpha=.0083 after Bonferroni correction.
The documentation times were not normally distributed (
On the basis of the ANOVA of documentation time, the model was statistically significant (adjusted
Document quality was assessed using 8 observed PDQI-9 metrics (
The usability data were analyzed using a paired
This formative study sought to assess the feasibility of using an EHR documentation method based on dictation and NLP by evaluating the effect of the method on documentation time, documentation quality, and usability. We found that a pure protocol of NLP Entry as well as hybrid protocols (involving both NLP Entry and Standard Entry) showed promise for EHR documentation, relative to Standard Entry alone (Standard-Standard Entry). It is our opinion that different parts of the note should be documented differently, but reaching a conclusion on the optimal method of documentation for each part of the note will require further study.
The finding that NLP-NLP Entry and NLP-Standard Entry required significantly less time than Standard-Standard Entry can be explained by the faster speed of dictation relative to that of entering data using the keyboard and mouse, rather than by the involvement of NLP.
No statistically significant difference was found between the overall documentation quality (measured using the PDQI-9 tool) of Standard-Standard Entry and that of any of the other 3 documentation protocols. The succinctness of Standard-Standard Entry documentation was found to be significantly greater than that of the other 3 protocols. This suggests that the note was judged to be more to the point and with less redundancy. In addition, documentation from Standard-Standard Entry was found to be more organized than that from NLP-NLP Entry, indicating that it was structured in a way that the reader could better understand the patient’s clinical course. When the participant used the Standard-Standard protocol, they used Standard Entry for history and physical examination sections as well as assessment and plan sections. When they used the Standard-NLP protocol, they used Standard Entry for history and physical examination sections and NLP Entry for assessment and plan sections. In the former (Standard-Standard), the participants tended to type shorter paragraphs for the assessment and plan sections. In the latter (Standard-NLP), they dictated the assessment and plan resulting in a larger volume of text. This difference warrants future scrutiny. On the basis of the results of the modified SUS, the participants’ usability ratings for NLP-NLP Entry were significantly higher than for Standard-Standard Entry. These findings suggest that, pending further study, EHR documentation methods using a combination of dictation and NLP show potential for reducing documentation time and increasing usability while maintaining documentation quality, relative to EHR documentation via standard keyboard-and-mouse entry.
Documentation methods using dictation and NLP have the potential to reduce some of the most egregious “pain points” for EHR data entry. These methods can facilitate capture and insertion of both structured data and transcribed text into the appropriate EHR sections, affording the user of the note the option of using one or both types of information. The structured data are ideal for interoperability and coding and may prove to be useful for analytics.
Opinion is divided regarding the relative advantages of narrative fields and structured fields in clinical documentation and in which contexts each excels or is preferable [
In future research, for the purpose of achieving documentation quality using dictation and NLP that, in all respects, is comparable to or better than documentation quality resulting from Standard Entry, certain changes to the NLP Entry process will be evaluated. We will assess the effects of requiring participants to use dictation under the constraints of a structured template on improving the organization, comprehensibility, succinctness, and synthesis of notes produced from NLP Entry. The templates would reflect the structure of the participant’s EHR. In addition, we will aim to improve the procedures by which NLP output data are translated into and transferred to the clinical note. We also plan to more systematically scrutinize data capture differences pertaining to documenting in different sections of the EHR note. This will enable us to fine-tune hybrid methods of data entry.
In a subsequent study, we will measure the time required for, and documentation quality and usability of, NLP Entry in live clinical use. This will require developing automated interfaces for sending the participants’ dictation to MediSapien NLP and for sending structured data and free text from MediSapien NLP into the EHR, during which process the participant will be able to modify the documentation. This process will be facilitated by the emergence and widespread adoption of interoperability standards and messages that can carry rich structured data.
This study has several limitations. One limitation is that the simulated interface used in this controlled experiment is somewhat lacking in ecological validity. In a real-world live setting, the structured data and the transcribed text data would both be inserted into the EHR via an automated interface. In addition, the physician would be able to review or modify the documentation before it was finalized. For the purposes of this formative study, this process was simplified. Therefore, an interface to automatically insert the structured data and text into the EHR or allow the physician to review the documentation before finalization was not used for this study. Instead, the insertion process was simulated by manually generating a note in Microsoft Word resembling one that might have been generated by the automated insertion process. Time required for generating the note was not included in the study’s time measurements. To ensure that the manually generated note could have been produced by an automated process, it was produced following strict predetermined rules and without any reliance on human discretion.
Second, physicians generated documentation for the study based on test scripts about fictitious patient encounters. Test scripts included history and physical examination sections and were formatted as transcription notes. The assessment and treatment plan sections were excluded from the test scripts. Participants were instructed not to dictate or type verbatim what was written in the test scripts, but to understand what was written and document it in their own way. In addition to being instructed to generate history and physical examination sections, they were instructed to generate their own assessment and treatment plan sections, because those sections were excluded from the test script.
The sample size for cardiology and nephrology was rather small owing to recruiting challenges. This affected the power for determining differences for related contrasts. Clearly, a larger sample size would have enabled us to detect more subtle group differences.
A limitation of this method of generating test documentation was that because it presented medical information in a free-text format, it may have favored documentation methods requiring the physician to generate free text. NLP Entry requires documentation via dictation exclusively, and Standard Entry entails only some documentation via typing, with the rest entered by pointing and clicking using a mouse. Consequently, the results for time required to complete documentation may be biased toward free text and therefore toward NLP Entry. Nevertheless, we perceive a value in measuring the temporal differences and think that such differences may be consequential in real-world use of this system.
Current standard methods of EHR documentation have been shown to be extremely time consuming and are judged to have suboptimal usability. In this formative study, the feasibility of an approach to EHR data capture involving applying NLP to transcribed dictation was demonstrated. This approach was shown to have the potential to reduce the time required for documentation and improve usability while maintaining documentation quality in several respects. Future research will evaluate the NLP-based EHR data capture approach in a live clinical setting where generated structured data and transcribed text for real patients are inserted into the EHR via an automated interface.
The past decade has witnessed a dramatic increase in the adoption of EHRs as central instruments in medical practice. However, these systems have not yet proven to be reliable tools for facilitating clinical workflow or enhancing patient care. Recent advances in usability have led to the development of frameworks, new methods, and robust assessment tools that can be used to more precisely delineate the source of the problems associated with an interface [
There is ample evidence that clinicians spend many hours documenting patient records and sometimes at the expense of time that could be devoted to patient care. Dictation is a familiar method of data entry to most clinicians. The proposed solution leverages that familiarity and has the potential to produce a quality document or patient note in less time along with highly structured machine-readable codes.
analysis of variance
computer-based documentation
Columbia University Medical Center
electronic health record
Health Information Technology for Economic and Clinical Health
Medical Language Extraction and Encoding System
natural language processing
Physician Documentation Quality Instrument
System Usability Scale
The authors thank the clinicians who participated in the study. The research was supported by the Center for Advanced Information Management of New York State (#13240001) and the National Library of Medicine of the National Institutes of Health under Award Number R43LM011165. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
JM, AB, and AF are employed by ZyDoc Medical Transcription LLC, whose MediSapien NLP software product was used in this study.