Usability and Safety in Electronic Medical Records Interface Design

Naturalness

All information in the display should appear in a natural order. Naturalness also refers to how familiar and easy an application is to use and to what extent it follows the “natural” workflow of the system. In the field study that Hollin, Griffin, and Kachnowski (2012) conducted, users indicated neutral-to-positive attitudes toward this aspect of EMR interfaces; however, authors of other studies have identified violations of naturalness in EMR interfaces. Harrington, Porch, Acosta, and Wilkens (2011) conducted an empirical study that used heuristic analysis to evaluate EMR designs in terms of 14 usability design principles. By monitoring staff nurses working with EMRs, they found that one of the most frequently violated usability principles was a lack of matching between user work flow in the real health care operations and using an EMR. They recommended improvements in EMR design based on the heuristic analysis but did not specify methods by which to achieve new designs. Zopf-Herling (2011) conducted a comprehensive analysis of EMR usability by reviewing information displays that were shared by hospitals. On the basis of this empirical study, the author offered that interface design should support effective communication of data. The author suggested logically ordering interface content based on work flow and highlighting task-critical information. Zopf-Herling also offered that considering user characteristics in EMR interface design could promote usability and encourage more frequent use of EMRs.

Beyond the aforementioned work, authors of other studies have identified violations of naturalness as part of EMR or EHR usability evaluation. For example, Pereira et al. (2012) found that the principle was violated in one EHR form. They recommended that the position of interface buttons be changed in order to achieve the natural work flow of actual hospital tasks. In another study, Craig and Farrell (2010) identified information recording and navigation as the two tasks suffering from usability issues in EMR systems. Focusing on information entry, they said that EMR interface design should make entry processes as natural as possible for physicians. In order to address this issue, they designed a new interface that allowed users to enter data in a free-text format (vs. using templates or selecting terminology/labels). The expected outcome was reduced disruptions between clinicians and patients during consultations. In addition, Schumacher, Berkowitz, Abramson, and Liebovitz (2010) found physicians to perceive mismatches between EHR work flow and actual clinical processes. They concluded that EHRs should not hinder physician performance but should allow them to work “smarter” and faster by leveraging real task work flows in design.

In their review of EMR technology, Harrington, Kennerly, and Johnson (2011) assessed performance with three components of EMRs, including CPOE systems, clinical decision support systems (CDSS), and bar-coded medication administration systems. Based on several studies, they concluded that CPOE systems could be a new source for errors in EMR use that might lead to adverse patient consequences or even mortality. They attributed the potential for such errors, in part, to poor EMR interface design. Some safety issues identified by Harrington, Kennerly, et al. included changes in work flow and task interruptions posed by EMRs. They suggested that effective planning and design of CPOE systems is critical for reducing unintended consequences. Finally, in a more recent study, Rogers, Sockolow, Bowles, Hand, and George (2013) identified violations of the naturalness principle as a part of a usability analysis of a nursing information system (NIS), which was one of the modules of an EHR system used in a hospital. In an empirical study of nurses in two hospitals, and application of a scenario-based evaluation technique (think-aloud) as well as usability heuristic evaluation, the authors found that NIS screen inputs did not match clinical practices. For example, the order of data entries was not based on the usual order during a physical exam. The researchers said this problem would lead to nurse frustration and reduced productivity.

Consistency

This principal basically means that knowing one part of an interface should be relevant for use of other parts. A particular system action should always be achievable by one particular user action (Molich & Nielson, 1990). More consistent systems also require less learning, which ultimately leads to fewer errors in task performance. Related to this definition of consistency, Dix et al.’s (2004) categorization also defined consistency as a subset of system learnability.

In our review, there were very few studies in which lack of consistency was identified as the main usability problem related to EMRs. Nakano and Tohyama (2009) evaluated an EMR system between 2005 and 2006 and focused on nonfunctional characteristics. Their results revealed a significant problem in lack of system design consistency, including colors, the meaning of colors, and layout. They also identified a lack of unified operating procedures and lack of consistency from one screen to another, even within one EMR product. The authors mentioned some causes of inconsistency in the EMR system, including various situations occurring in the development process, local design rules, and requirements for specific additional functions. In order to prevent these errors, they provided recommendations in three major categories of unified design, unified operation, and improved operability. Some of their specific recommendations included using specific basic colors and unified layouts of forms, applying unified design to buttons that have similar functions, and designing entry areas based on emphasis and simplification principles. After formulating a design guideline, they were able to upgrade their EMR system and eliminate the inconsistency problems.

Edwards, Moloney, Jacko, and Sainfort (2008) also identified violation of the consistency principle in their usability evaluation of a commercial EHR in a pediatric hospital system. They used the heuristic walk-through method to evaluate the system in a case study. Users of the EHR system consisted of physicians, nurses, pharmacists, therapists, other clinicians, and support staff. Results showed that 15% of issues were related to consistency. For example, they found that the close button was not always in a consistent place. However, they provided some general recommendations to address usability issues, which included immediate changes in system configurations and training material, and sharing evaluation results with vendors in order to improve usability in future releases.

Some other authors mentioned a lack of consistency as part of other EMR design problems. For example, Weir et al. (2003) reported that inconsistent text format requirements (e.g., “mm/dd/yyyy”) was an issue leading to frequent data entry errors in EMR use. In addition, they suggested that since EMRs are typically reviewed by physicians using a “skimming” process,” the interfaces should be designed in a consistent and familiar format in order to reduce learnability and increase speed of task performance. However, Hollin et al. (2012) found that users have neutral-to-positive attitudes toward consistency in EMR design.

Preventing Errors

Interactive system interfaces should be designed in a way that prevents errors from happening in the first place (Molich & Nielson, 1990). Schumacher et al. (2010) observed that EMR interfaces should anticipate the next steps of users and prevent potential mistakes in task performance. Related to this observation, Edwards et al. (2008) found that users may not know what to do in the next step. For example, there was a required data entry field for which users did not know the value to enter at a specific stage of system use. This situation resulted in data entry errors. To address the problem, the authors recommended changing the required field to a nonrequired field or instituting a policy on how the system should respond when the user is uncertain about a required value. Related to this problem, Rogers et al. (2013) mentioned that some functions of the NIS were unclear to nurses and they could not predict what would happen if they clicked them. To address this issue, Rogers et al. recommended that an NIS should support the ability for users to predict when memory of information would be required in EMR work and to provide prompts to cue data entry.

In our review of literature, there were several studies in which authors identified violations of the error prevention principle in providing copying and pasting (C/P) functions for data entry as part of the documentation process. For example, Weir et al. (2003) analyzed notes made in EMRs as part of inpatient admissions and observed that C/P functions, object insertion, and electronic signatures were the main causes of data entry errors. To solve these problems and prevent future errors, they recommended enhancements to EMR design, including structured C/P functions and reducing the use of templates for data entry. Regarding C/P function use, Thielke, Hammond, and Helbig (2007) explored the prevalence of use in EMRs and the effect on medical errors by using an automated text categorization algorithm. They conducted an analytical study of patient records and found C/P of exam results to increase errors in records. The authors recommended altogether removing C/P functions from EMR software or presenting copied and noncopied texts in different colors for user awareness.

Related to this finding, O’Donnell et al. (2009) also noted inconsistencies and outdated information as a result of C/P function use in patient notes. They conducted a cross-sectional survey of resident and faculty physicians in two medical centers. Although most of the physicians identified errors from using C/P functions, the majority of them wanted to continue to use the functions. To resolve this issue, some users agreed with recommendations, such as having C/P text more identifiable (highlighted or italicized), adding alerts to notify users when notes are similar, and placing restrictions on C/P of parts of notes (e.g., physical exam information). The use of C/P functions resulting in EMR content errors was also one of the interface design problems identified by Caudill-Slosberg and Weeks (2005). They also said that ensuring the availability of readable documents as well as elimination of C/P functions would reduce the distribution of inaccurate data and improve overall patient safety.

In a more recent study, Weis and Levy (2014) identified benefits and risks of content-importing technologies (CIT), such as C/P functions in EHRs. Despite CIT advantages, such as saving time, these technologies may cause risks that compromise patient safety. In their review of literature, Weis and Levy found CIT tools to lead to inconsistency and repetition of expired information and to create difficulties for users in identifying new information. To address these problems and prevent errors, the authors summarized controls from the literature, including copying information from one record to another, copying chief complaints, reviewing systems and physical examination information, copying entire sections of documents and reports, copying subjective data, and so on. They also provided a list of “best practices” for EHR interface design. They recommended clearly identifying all copied information and including attributions to sources of information (e.g., date, time, and original author) in copied text or data.

Hyman, Laire, Redmond, and Kaplan (2012) identified errors in placement of orders in patient EMRs. They observed that orders were often made for the wrong patient or the wrong order was made for a patient. In order to address this problem, they made changes to a current EMR interface used in a large hospital, including an order verification dialog with a patient photograph. This dialog was presented directly before final electronic signature of orders. They compared the number of reported incidents in care between the existing and new systems and found that order verification significantly reduced the frequency of unintended documentation in patient EMRs. As with orders being incorrectly filed for patients, Wilcox, Chen, and Hripcsak (2011) identified “patient–note mismatch” as a common type of EMR data entry error. They observed that documents were often added to EMRs without confirmation of patient names and numbers and so on. They designed a new interface, adding a pop-up window to show patient identification before confirmation of a new record submission. In this way, the interface interrupted user action and required review of patient identification before a system upload. By comparing the rate of patient–note mismatches before and after the new intervention, they found that the error reduction rate was significant. The authors also observed the interface to make correct patient information more accessible to the user.

Shachak, Hadas-Dayagi, Ziv, and Reis (2009) also observed that EMR use can lead to additional medical errors, such as patient–note mismatch and typos. By conducting a cognitive task analysis using semistructured interviews, as well as field observations of primary care physicians, they found system automaticity to be the main factor that could lead to errors. Although their findings indicated a fine balance between the benefits and risks of EMR use, on the negative side, they found that users accidentally added information to the wrong patient’s chart and selected wrong items from scroll-down lists located above or below desired items. They also mentioned that when a list of patient names appeared by typing part of a name, physicians unintentionally selected the wrong name. The authors said that providing the capability for physicians to double-check prescriptions before signing them would serve as a safety mechanism in order to prevent such errors.

In a more recent study, Yamamoto (2014) conducted a survey of emergency department clinicians to assess the frequency of errors in patient notes in EMRs. They found that almost all clinicians made wrong patient EMR errors. In order to prevent these errors, they improved the information display by adding patient room numbers as watermarks in an EMR to display, with automatic and passive presentation. The author said that other methods to prevent patient–note mismatch, such as adding a confirmation page (e.g., Wilcox et al., 2011), would result in additional burden on clinicians. By displaying patient identifiers (such as room number) as watermarks, EMRs can prevent errors in a passive and more automated way.

Bowman (2013), in studying EHRs, observed that the complexity of interfaces contributes to user confusion and can cause errors. In her literature review, she identified the main causes of documentation errors to include C/P function use, use of templates, use of standard phrases and paragraphs, and automatic insertion. With respect to templates, Bowman observed that default data entries or prior field fill-ins led to identification of patient characteristics not representative of actual conditions. She also observed a lack of consistency in the design of data fields from one template to another, making detection of such default entry errors difficult for users. Beyond this finding, Bowman found that errors were often committed in template use as a result of adjacency of features and fields (i.e., confusing one field for another). The author provided recommendations to address these issues, such as monitoring the C/P process to ensure appropriate case content and implementing EHR content design standards and guidelines.

In the most recent study, Lowry et al. (2014) used two human factors modeling methods (process mapping and goal–means decomposition diagrams) to improve EHR work flow integration in a larger clinical work flow. They considered using EHRs in different stages of the health care process (before the patient visit, during the visit, physician encounters, discharge, documentation). In the physician encounters and documentation stages of the process, physicians identified several usability issues regarding the use of EHRs. These issues included reduced user vigilance and increased error rates in use of many C/P functions, use of many templates for data entry, and inaccurate or difficult modification of diagnosis lists. The recommendations they suggested to address these problems included the use of color-coding methods to differentiate between copied and new information, providing the capability to label diagnoses with degrees-of-certainty information, and providing an unknown option to be selected from a diagnosis list.

Finally, Méndez and Ren (2012) found that, in general, reading and entering, for example, the vital signs of patients in an EMR is an error-prone and time consuming process. To address this issue, they proposed a prototype design of an electronic interface for the EMR system to automatically acquire and store data for other networked systems and databases. In our review of literature, there were other studies in which authors found violations of error prevention as part of their analysis. For example, Abramson et al. (2012) found that having a default value for dosage fields is error prone (especially for residents) because users often accept computer-generated doses without checking accuracy. In addition, Hagstedt, Rudebeck, and Petersson (2011) found that entering dosage values as free text may result in different interpretations at a pharmacy and ultimately increase the risk that a drug will be prescribed with the incorrect dosage. The authors said that an improvement in the dosage function of CPOEs is necessary.

Minimizing Cognitive Load

Human short-term memory is limited in capacity (Miller, 1956). Interfaces should be designed in a way to reduce mental workload for users. Users should not have to memorize system information or database content (Molich & Neilsen, 1990). Information overload is a problem that occurs when perceptuo-cognitive capacity is exceeded by the quantity of data presented via an interface to the extent that errors occur in user information processing. Authors of several studies we reviewed identified violations of the minimization-of-cognitive-load principle in EMR interface design. For example, Kuqi, Eveleigh, Holzer, and Sarkani (2012) evaluated usability issues in EMR use and observed that poor design leads to user information overload. They proposed an approach to analyze dependencies among display information elements called the design structure matrix (DSM). They applied the DSM to a case study of ambulatory-based EMR system use and formulated interface design changes that minimized the number of steps to complete a task. The authors offered that the reduced task steps served to minimize user cognitive load. In another study, Ahmed, Chandra, Herasevich, Gajic, and Pickering (2011) identified user errors in EMR use due to information load and prototyped a new interface design that ranked data in terms of importance to a case and presented only highly ranked items on the screen for clinicians. The authors compared the new interface design to a standard design and measured information processing load using the NASA Task Load Index. Through this empirical study, Ahmed et al. found the new interface to reduce errors and task completion time and to facilitate efficient information navigation.

Through an analytical study, Harrington, Wood, et al. (2011) also identified information overload as a major EMR usability issue. They observed that EMRs typically include many objects on screen and that several may not be relevant to a user’s work domain. They found that such designs create (cognitive) “overhead” for users and that the amount of domain-related content needs to be increased in interfaces. They also found that reductions in the overall information content and complexity of EMRs could promote task efficiency and ease of use. They redesigned a system interface with the goal of reducing user cognitive processing. The redesign reduced the number of screens and dialogs used to present similar information. They also modified “confusing” dialog boxes by limiting options to those considered “important” to case processing and by removing “unimportant” options. Widgets were, however, provided to guide users to hidden information. They also suggested that direct manipulation methods for modifying interface content could lead to reductions in the number of steps and time for performing tasks. They employed the task analysis, user analysis, representational analysis, and functional analysis (TURF) method to evaluate the usability of the existing interface and redesign and found the redesign to achieve the design goal.

In a more recent study, Suebnukarn et al. (2013) found mental workload to be a major usability issue in using EHRs. They analyzed an EHR interface used in a comprehensive dental clinic by applying the cognitive task analysis method “goals, operators, methods, and selection rules” as well as keystroke level modeling. Results revealed mental operations to be the majority of steps in interface use, and more than half of user time was spent on performing mental tasks, which can result in fatigue over the long run. The authors wrote that designers can simplify interfaces and improve organization by combining related information and using fewer screens. In addition, the authors mentioned that providing information reminders and recognition-based assistance can reduce mental workload. Other suggestions included use of different colors to classify each main menu in order to reduce mental recall and improve user-friendliness of interfaces.

Cognitive overload has also been identified as an EMR usability problem in evaluations conducted as part of some other studies. For example, Rose et al. (2005) suggested that when the amount of information presented in an interface increases, user attention is divided, as in multiple-task performance, and the probability of errors increases. They identified “close” screen elements as an example interface design problem. To address this issue, they recommended presenting related items near each other (based on the proximity-compatibility principle described by Wickens and Carswell, 1995, according to which displays with relevance to a common task or mental operation should be rendered close together in perceptual space). Rose et al. also suggested presenting only certain test results on displays rather than presenting all notes. In addition, Rogers et al. (2013) found that most users felt they had to remember actions already completed as well as what they had to do next, which the authors contended causes a high memory load. To address the problem, they recommended making status information more visible in the NIS module of EHRs.

In another study, Schumacher et al. (2010) assessed EHR usability, including interface information content. They conducted an empirical study, including interviews with physicians, in order to identify usability problems and potential solutions. They also found EHR content to cause information overload for physicians as well as “alert fatigue.” Related to this finding, they concluded that EMRs should present essential information and that there is a need to eliminate irrelevant interface options. Alerting overload was also a usability and safety issue identified by Bouamrane and Mair (2013), Abramson et al. (2012), and Sittig and Singh (2012). Last but not least, minimizing cognitive overload was also identified as an EMR design rule of thumb by Zopf-Herling (2011). The author suggested reducing redundant information, reducing the number of dialog options, providing information summary screens, and carrying critical patient data forward from one mode of operation to another in order to address cognitive overload problems.

Although authors of a number of prior studies have recommended reducing the amount of information presented in EMR interfaces, Weber-Jahnke and Mason-Blakley (2012) observed that some EMR screens did not present critical patient information, including current infusions or identification of current lab test results. They applied a systems safety analysis approach (reviewed later) in order to identify hazards or safety issues associated with EMR design. However, they did not provide recommendations to solve the specific information presentation problems. Finally, in Hollin et al.’s (2012) study, the authors found that users have neutral-to-positive attitudes about the level of cognitive load imposed by EMR interfaces.

Efficient Interaction

Human–computer interaction should be designed for efficiency by minimizing the number of steps to complete a task or providing shortcuts for users (Belden et al., 2009). Molich and Nielsen (1990) recommended using shortcuts in order to make the system easy to use for all users. Belden et al. (2009) also identified reducing the amount of visual search for information and minimizing cursor travel distance as other objectives of efficient interaction.

Craig and Farrell (2010) focused on assessment of EMR interface navigation, function accessibility and flexibility, and recording patient information. They observed existing interface designs to inflate the number of steps to perform a task and that displays failed to maintain an overview of information for users. They also noted the presence of many “modal” dialogs, or interfaces requiring a user response before any other system information could be accessed. They presented a new interface design by reducing conventional interface elements to improve the navigation process and physicians’ work flow. However, at the time of their publication, the EMR prototype was under development and usability evaluation had yet to be conducted. Related to Craig and Farrell’s study, Saitwal, Feng, Walji, Patel, and Zhang (2010) examined the usability of the Armed Forces Health Longitudinal Technology Application EHR system by using cognitive task analysis. They found the current interface design to require a large number of steps to complete tasks. They argued that this situation represented an interface navigation problem and that it posed high mental demands for operators with the potential for information overload. The authors recommended reducing the number of steps and menus required to perform system functions and to ensure that interfaces presented task-critical information to reduce the level of difficulty for users.

Minimizing the number of user control actions to access functions was also one of the usability rules of thumb identified by Zopf-Herling (2011). She also suggested the use of check boxes instead of drop-down boxes in order to reduce the number of clicks for completing a task. Related to this finding, Hagstedt et al. (2011) compared the usability of different CPOEs used in EHRs in primary care. By developing an evaluation model and conducting interviews with physicians, they found that one of the most prominent issues was nonintuitive interface design. They observed that many mouse clicks were required to make a prescription. Although some of the required steps were mandatory due to patient safety issues, the authors wrote that a balance should be achieved between simplicity and patient safety and that more intuitive systems should be developed.

Zheng, Padman, and Johnson (2007) conducted an analytical study of the organization of EMR content and the potential impact on user task efficiency by using sequential pattern analysis. They analyzed user activity sequences with EMRs in order to identify patterns in navigation of features. On the basis of this analysis, they found that many features with high frequencies of use were not prominently located. Conversely, they noted that less frequently used items often occupied salient interface spaces. They also observed that many related information items were not colocated or adjacent to each other. They recommended that EMR features be relocated based on frequency of use and organized according to functional relationships. They also observed that usable EMR interface design could be facilitated by analyzing actual user everyday usage patterns for inefficiency.

Rose et al. (2005) sought to identify usability deficiencies in EMR design, including feature navigation. The authors conducted two qualitative empirical studies, including a user task analysis and ethnographic observations (patterns of group behavior with interfaces). The studies were focused on primary care practices and revealed that user alerts and process guideline dialogs did not appear at prominent EMR screen positions. They also observed that accessibility to EMR functions could be substantially improved by adding function controls presented as graphics at the bottom of display screens. Related to this finding, Rodriguez, Borges, Murillo, Ortiz, and Sands (2002) compared a graphics-based electronic patient record (EPR) with a text-based record in terms of the usability attributes of learnability, task efficiency, and user satisfaction. They conducted a lab study and asked users (internal medicine resident physicians) to perform the same tasks with both interfaces. They recorded performance times and found the average time for completing tasks to be significantly less when using the graphics-based interface prototype; that is, the efficiency of interaction was greater as compared with the text-based interface.

Jaspers, Peute, Lauteslager, and Bakker (2008) compared a newer version of an EMR system with a previous one in order to understand if the redesigned system improved the satisfaction of users. The authors administered usability questionnaires to clinicians. Results showed that although users were generally satisfied with the new EMR system, screen design and navigational structure were less easy to work with. Regarding navigational issues, the authors said that there was too much information on a single screen, which made it difficult for users to complete tasks. They recommended that EMR systems should be designed in a consistent manner, functions should have one-to-one correspondence with user goals, and the order of information presentation should match the order of information processing in a user’s mind. These recommendations are also in line with the design guidelines for minimizing cognitive load.

Indranoi, Wyatt, and Li (2009) developed an EHR training tool called iCare and assessed tool usability by applying Molich and Nielsen’s (1990) heuristic analysis method. Nursing faculties and students evaluated the interface based on questionnaires. The main usability issue was related to use of the keyboard. The authors found that the user interface mainly consisted of check boxes and radio buttons. They believed not having free-text boxes presented a usability issue. In order to solve this problem, the authors added a check box labeled other. When the check box was selected, a text box appeared for entry of free-text values.

Abramson et al. (2012) also found violations of efficient interaction in their comparison of new versus old versions of an EHR system. In a case study, including field observations and semistructured interviews of internal medicine faculty members, they identified usability as one of six major design issues. Related to this finding, the authors found that most physicians described the new EHR as “overengineered.” They mentioned that the new interface supported too many ways for doing tasks, which was confusing for users. In addition, there were many steps to complete tasks, and it was difficult for users to find certain information. Related to this issue, the authors recommended simplifying the interface and its functions. They also mentioned that if additional steps were required in use for safety purposes, real-time feedback should be provided in order for users to understand the advantages of these new steps.

Bouamrane and Mair (2013) assessed general practitioners’ (GP) ideas on using an EMR system and found navigation to be one of the main usability issues. They conducted semistructured and open-ended interviews to collect GP views on information management in a patient surgical flow in a major medical system in Scotland. Although many practitioners believed that the EMR system was beneficial to their work, they expressed some usability concerns, including unnecessary steps (a lot of clicking), extra keystrokes, a busy “front page,” and difficulty in visualizing the flow of patient consultation through the system. They recommended that iterative user-centered improvements and additional training in using the technology would solve these issues.

In a more recent study, Vatnøy, Vabo, and Fossum (2014) also identified navigation as a main usability challenge of EHRs. The authors conducted an empirical study with registered nurses and used the cognitive walk-through approach to identify usability issues related to the graphical user interface of an EHR. Results revealed finding information where it was expected to be found was rated low by participants. In addition, users could not find the information easily when they were looking for it. This situation was primarily due to functions not being organized in a logical way. Another issue with navigation of the system was that it was not intuitive where information could be found, and none of the participants were certain as to where to look next. The authors recommended that education, training, and support are necessary for using EHR systems. In addition, they offered that standardization of format, content, and terminology is needed to improve EHR functionalities.

Violation of the efficient interaction principle was also identified as a problem in some other usability assessments of EMR interfaces. For example, although Walji et al. (2013) primarily focused on display terminology problems (discussed later), they also found other navigation problems, such as limited accessibility of functions or time-consuming processes to select functions, illogical organization of options and terms, and inconsistent positioning of controls for functions. In another study, Schumacher et al. (2010) also observed interface navigation problems, specifically, excessive clicking to access functions, resulting in physician frustration. Rogers et al. (2013) said that nurses also complained about navigation issues in the NIS module of EHRs. However, their recommendation, which was increasing observability of work processes, did not directly address the usability problem. Last but not least, Edwards et al. (2008) said that users had to reenter the same information for each new order, which violates the principle of efficiency in task completion. To address this issue, the authors recommended that the system be designed in a way to inherit information from previously completed forms.

Forgiveness and Feedback

Molich and Nielson (1990) wrote that interactive systems should provide feedback in real time in order to keep the user informed about what is currently going on. Appropriate feedback should also inform users about the consequence of actions they are going to make (Belden et al., 2009). In general, feedback is critical for facilitating user error recovery and presenting future errors. Related to this issue, Ventura et al. (2011) suggested that EMR systems should be equipped with CPOE dialogs in order to increase efficiency in use; however, they also noted that without patient and order verification dialogs, such systems could lead to prescription errors. After an analysis of the data from a 10-year EMR usage study, the authors’ main conclusion was that the success of EMRs is highly dependent on user involvement in the system design and development phases for conveying feedback requirements. With respect to their specific EMR interface design, Ventura et al. recommended integration of an “alarm” into the system to inform users whenever an erroneous entry was made in a prescription.

No clear feedback or delayed feedback on an action are usability issues that were also identified in several other studies (e.g., Pereira et al. 2012; Weber-Jahnke & Mason-Blakley, 2012). Graham et al. (2008) offered that when there is no feedback to identify incomplete order entries (under high workload conditions), users often fail to recognize the oversight. In a more recent study, Rogers et al. (2013) identified lack of feedback as a violation of visibility of system status. They said that because of no feedback in the NIS module, users did not know if others saw their documentation. In addition, users had no idea where the data went in the system after they entered it. Their recommendation regarding this issue was to provide status information to users, especially making system changes more visible.

Effective Use of Language

Molich and Nielson (1990) wrote that all dialogs should be presented with clear words and phrases that are familiar to users. In the health care domain, there are many terms and abbreviations that may be familiar to specific users but may be meaningless to others. Belden et al. (2009) also mentioned that abbreviations and acronyms should be used in EMR interfaces only when they are completely understandable and meaningful to all users.

Walji et al. (2013) evaluated the usability of an EHR interface in order to identify some high-priority problems that could deteriorate user ability to perform tasks. They conducted a field study that included user testing, interviews, and direct observations from dental providers for problem identification. They also focused on the organization and identification of data fields and options in the interface. They observed that confusing labeling of controls and excessive use of abbreviations created the potential for user errors. They found that the terms used in the EHR interface were not optimal for users, and they had difficulty in understanding the meaning of some categories and concepts. They concluded that HCW success in use of EHRs is critically dependent upon interface usability.

Graham et al. (2008) also studied usability issues of CDSS, including data field design and entry. A CDSS is any electronic or nonelectronic system that helps decision making and uses specific information to generate assessments or recommendations that can be presented to clinicians for their consideration. The methods Graham et al. used to identify usability issues included automated screen captures combined with analysis based on usability engineering techniques. They observed, for example, that the absence of desired terminology in menus or dialogs often led to user selection of similar options but not exactly correct choices (in order to avoid leaving fields blank). They said that the identified usability problems could lead to adverse medical events. Graham et al. also suggested that usability problems need to be solved early in the system design phase of the product life cycle in order to be cost-effective and to prevent the potential for negative patient outcomes.

Weber-Jahnke and Mason-Blakley (2012) suggested that deteriorating quality of terminology and data in EMR documentation represented a hazard for patients. By analyzing a case study of EMR safety issues, they observed manual entry of text, in the absence of user menu-based selection of terminology, to increase data entry and documentation errors. They found that when users could not find appropriate terminology in menus, manual entry was error prone. This situation creates data quality problems over time, such as misspellings of terms or the use of incorrect terms for medical conditions, depending on the background of the health care worker. Finally, they suggested that without embedded system controls to prevent or account for such data entry errors, a general reduction in EMR data quality and patient safety would ultimately occur. There is a need to provide HCWs with medical terminology reference tools in the use of EMRs.

Finally, Sittig and Singh (2012) also identified the need for structured and coded data in EHRs. On the basis of a literature review, they proposed a three-phase framework for development of EHR patient safety goals. In addition to alert fatigue, the authors identified a lack of structured data as contributing to failure in appropriate use of EHRs and creating safety concerns. The authors argued that this issue might prevent users from correct interpretation of system states, which could lead to errors. Some interface design recommendations for solving this problem included use of structured clinical vocabularies and developing order entry templates. Authors of other studies also referred to violations of the effective use of language principle in EMR design as a part of their usability evaluation (e.g., Indranoi et al., 2009; Vatnøy et al., 2014; Zopf-Herling, 2011).

Effective Information Presentation

The design of EMR interfaces, in terms of the amount, type, and organization of information, influences complexity and usability from a user perspective. Caudill-Slosberg and Weeks (2005) observed that interface templates are typically used in EMRs for standardization and elimination of extra details. Although templates may have been developed to improve user efficiency, the authors found that such interfaces often prevent users from recognizing most relevant pieces of information for a specific task. In some cases, templates were observed to create visual barriers to search for relevant information. Caudill-Slosberg and Weeks conducted an analytical study (using a fishbone analysis technique) and showed that template use could lead to errors in medication orders, inaccurate HCW interpretation of orders, drug overdoses, and so on. They recommended that the design of templates account for the system understanding of different users in order to enhance patient safety.

Readability was another aspect of effective information presentation that Belden et al. (2009) identified. Lack of visibility of options in EMR interfaces was also identified as a usability problem by Walji et al. (2013). Wu, Orr, Chignell, and Straus (2008) observed HCWs to have difficulty in reading information from EMR screens and in data entry, specifically with mobile systems. They conducted qualitative evaluations of HCW performance and found that display visibility and creation of new records were the most frequent challenges with mobile EMRs. However, the recommendations that they provided basically address customization and flexibility issues. They said that in order to eliminate user errors and increase EMR usability, HCWs need to be provided with flexible, fast, and time-saving methods. They recommended integrating new technologies in EMR systems, such as speech recognition. Related to visibility of options, usability issues regarding the size of letters, text, and buttons were identified by Vatnøy et al. (2014).

Tasa, Ozcan, Yantac, and Unluer (2008) suggested that icons be used in EMRs to promote user readability and task efficiency. They conducted an empirical study and administered questionnaires to hospital users in order to identify relevant metaphors for various health care operations. They wrote that previous studies had focused on user “mental models” for overall interface design and navigation issues but not on metaphors for information presentation. They designed icons for EMR interfaces and found prototype interfaces to improve readability. In line with the larger body of human–computer interaction research, they concluded that icon-based design is preferable to textual interface design.

Violation of effective information presentation has been identified as a usability issue as part of some other studies. More specifically, Rose et al. (2005) identified screen contrast as an important feature of EMR user interface designs. They recommended that screens be designed with higher resolutions and to incorporate soft colors to improve feature contrast for promoting readability. In another study, Edwards et al. (2008) found that 13% of users did not notice how to complete the task based on information presented to them. For example, some hyperlink texts were not underlined or highlighted in different colors, or some clickable icons did not appear as buttons. Jaspers et al. (2008) also found that the layout of information presentation in a new EMR system was less usable in comparison to an old version. They said that reorganization of information was somewhat more difficult to work with for clinicians. In order to solve this issue, they recommended clustering related information on the screen, which can also reduce mental load of users.

Customizability/Flexibility

Customization is the capability of an EMR interface to be modified based on the needs of each health care provider. Flexibility, or the capacity of an interactive system to be customized, is one of the 14 usability principles identified by Zhang and Walji (2011) in their research toward developing a unified framework of EMR usability. However, there are few studies in which customization is identified as an EMR usability issue. As previously mentioned, a prior report by the Health Information and Management Systems Society (Belden et al., 2009) did not include customization as a principle for EMR usability evaluation.

Chen and Akay (2011) identified a lack of interface customizability as a major issue in EMR system use in developing countries. To address this problem, they proposed use of database software (FileMaker) to allow clinics to easily change the information content delivered to an EMR interface based on specific needs. Rose et al. (2005) also identified use of predefined interface templates as a lack-of-customization issue in EMR design. They observed that many physicians prefer to use a free format in preparing letters and that templates did not support such preference. The literature reveals a need to provide flexible templates that allow users to import test results and letters in various formats in order to address diagnosis problems.

Pereira et al. (2012) identified lack of customization for users with different levels of expertise as a usability issue. They conducted a usability evaluation of an EHR in a Portuguese hospital by applying a usability inspection method and heuristic walk-through. They noted a lack of customizability for expert users, such as the amount of information revealed in notifications and error messages as well as access to shortcuts (e.g., hot-key sequences) to accelerate task performance. On the basis of this evaluation, the authors concluded that there is a need for new methods of information organization as well as control design to improve EHR usability. For example, Pereira et al. recommended adding “accelerators” to EHR systems to increase flexibility of use for different users.

In an empirical investigation of physician and nurse practices in recording information with EPRs, Reuss, Naef, Keller, and Norrie (2007) found the majority of HCWs to identify a lack of flexible ways for entering information to be a usability issue. They conducted interviews with the HCWs and found a desire for entry methods, including drawing sketches, annotating documents, or highlighting relevant data fields. They also noted that most of physicians and nurses could not record complete information in records due to the absence of suitable forms and elements, such as data entry fields that were very small. Based on these findings, they recommended that a shortcut and/or context menu should be provided for record annotation (marking specific information) and that any annotations should be displayed at the same time as the original EPR in order to increase system usability (efficiency of use). They also suggested that HCWs be supported in making independent entries and that all messages, questions, and so on should be managed in terms of the HCW’s role.

Lack of customization based on user’s role was also a problem identified by Lowry et al. (2014). They posited that EHR users want to see different views of patient information based on their role. For example, the data that are important for a physician might not be critical for a specialist, and the physician may prefer other information to be displayed on the screen. Lowry et al. recommended the use of multiple views for different physicians based on their role and adding a feature for viewing notes from the perspectives of other roles in order to ensure entered information meets the needs of other users.

Last, authors of some other studies mentioned an inability to reorganize interface content as a lack of customization. For example, Walji et al. (2013) observed reordering of concepts in an EHR to be a problem in their usability evaluation.

Design Guidelines

The foregoing literature review leads to a set of EMR interface design guidelines for promoting usability. These guidelines are listed in Table 5 along with the specific references identified as bases for each guideline. In addition, the design guidelines are grouped according to the classes of usability problems previously identified at the outset of the review.

Interface Design Concept

The guidelines in Table 5 provide a basis for an enhanced EMR interface design concept. Although the enhanced design concept is applicable to different health care clinics, we present the concept in the context of a physiotherapy clinic in order to make clear design features and how those features support specific processes. Most current EMR interfaces include a collection of electronic forms (designed based on paper predecessors) that are linked to a main or system “home” page. A physician or other HCW simply sees digital versions of forms filled with data fields that are to be completed either through free-text entry or by choosing from options in a drop-down menu list. Unfortunately, as noted from the literature review, this general design concept has led to usability problems, including information overload, physician entry of unclear terminology (i.e., violations of effective use of language), errors in C/P incorrect information (violations of error prevention), complex interface navigation (violations of efficient interaction), and a lack of customization for various levels of users. In general, we propose that EMRs be designed for specific clinic or health care provider use; that is, each provider should be presented with a set of screens or information that is directly relevant to its specific operations. Customized icons and provider- or clinic-specific terminology should be also used in the interface. Such an approach would be aimed at addressing the lack of customizability of existing interfaces.

Clinical use of EMRs is largely focused on patient problem description and prescriptions. A clinician typically logs in to a system using one screen and then views a patient’s records. In the enhanced EMR interface design, we recommend that the first information screen provide an overview of the system content using icons to represent various clinical functions. This design can address some of the usability problems identified in the literature, such as lack of effective information presentation. In the event that a clinician is creating a new patient record, there is a need for data entry and visualization. It is typical in EMR design for this function to be supported through use of templates; however, templates have previously been found to cause errors and user frustration (Caudill-Slosberg & Weeks, 2005) and to violate some usability principles, such as effective information presentation as well as prevention of errors. One approach to minimizing template use in EMR design is to stage user interaction with the system, that is, to support data entry with one dialog and form visualization through another screen or window. For example, in physiotherapy clinics, therapists may enter and select data using a dialog and view a complete form or record on another screen versus entering data directly into fields of the form. The advantage of the staged interaction is reduced user attentional load and decreased potential for errors.

With respect to data entry dialogs, icons or check boxes can also be used to represent different parts of the body or physiological functions that may be affected. Use of check boxes reduces the number of free-text format sections that users have to fill out. This change will lead to reduction in data entry errors and improve efficient interaction between users and the system. A therapist can select icons/options to pinpoint problem areas and simultaneously generate record content. Once a patient has specified an area of pain, graphical controls (e.g., slider bars) can be used to collect data on the severity and frequency of pain occurrences. Again, data entry using such controls can serve to generate required form/record content.

Depending upon the icons/options or ratings selected by a therapist at each stage of patient problem description, a set of potential disease states should be presented for clinician selection. By using this approach, the interface content can be reduced and the clinician can narrow down the set of possible diagnoses. This feature addresses the potential for user information overload problems, as previously identified in the literature. Once a diagnosis is selected, the interface should support clinician identification of potentially causal factors (e.g., occupational and/or nonoccupational) in the disease state. For example, a patient may identify severe low-back pain, and the EMR should support physiotherapist identification of posture, task repetition, force, and so on as potential work factors in the condition. We also recommend that this stage of the problem description be facilitated through icon selection in EMR dialogs in order to improve effective information presentation.

Finally, on the basis of the patient diagnosis and causal factor identification, prescriptions should be supported using separate dialogs with medication option selection and dosages. In the event that a special medication or dosage is needed, dialog design should include free-format text entry (Craig & Farrell, 2010) for the clinician. Although this approach can be error prone, the enhanced EMR interface design should include spell-checking and autocorrect functions to ensure accuracy of entries and prevent potential errors.

The main advantages of this new interface concept over current EMR designs include the following:

On the basis of the literature review, it is also important to observe a number of limitations of the existing body of research identifying EMR design problems, including the application of various methods of usability and safety analysis as well as the recommendations for design improvements. The majority of studies of EMR usability issues have employed descriptive or qualitative analysis methods, such as task analysis, functional analysis, user observation, user testing, interviews, behavior pattern analysis, heuristic evaluation, records review, and evaluation of reports before and after EMR implementation. Harrington, Kennerly, et al. (2011) made a similar observation in their review of literature on studies of EMR safety issues, specifically saying that descriptive or qualitative analysis methods had been used along with case studies and comparisons of designs. Related to this finding, the majority of studies identifying EMR usability issues make comparisons only between paper-based systems and EMRs. There are very few studies that include comparisons among multiple EMR designs in order to identify advantages of one alternative over another. Furthermore, to date, there are few studies presenting comparisons of different versions of an EMR interface in terms of the dimensions of usability (i.e., effectiveness, efficiency, satisfaction).

Another limitation of the present body of research is that among the studies making use of safety hazard analysis techniques to evaluate EMRs, no authors evaluated specific aspects of interface design. Authors of the studies reviewed here identified only general hazards related to the safety of EMRs, and system interfaces were identified only as an element that may lead to patient safety issues. Although these analyses are valuable in terms of providing general EMR safety guidelines and recommendations for the health care industry, they may not be useful in terms of controlling specific hazards related to EMR interface design.

Beyond this, all the hazard analyses presented in the previous studies employed inductive reasoning methods, including FMEA, TRA, STAMP, fishbone diagrams, and cause-and-effect analysis. That is, the methods allowed the authors to identify sources and mechanisms of hazard exposure and project potential negative outcomes. Although inductive safety analysis techniques appear to have been useful for identifying hazards associated with EMR use, there is a need to apply additional types of analyses, including deductive reasoning approaches, that is, hypothesizing negative patient events and deducing potential sources and mechanisms in EMR systems. Despite Weber-Jahnke and Mason-Blakley’s (2012) contention that fault tree analysis (FTA) is not applicable to EMR use, the majority of failures in health care involving EMRs are attributable to specific events (e.g., HCW data entry) and occur in specific use scenarios. Fault trees can be created for specific EMR failure modes and contexts of use. It is also possible to develop a collection of fault trees (organized according to an event tree/scenario) for objective quantification of the frequency of potential hazards as well as identification of severity of outcomes for patients. The use of such methods may reveal additional causes of EMR failures and negative patient outcomes.

Based on these limitations, there are several directions of future research that should be pursued, including the following: