E-Iatrogenesis and EHR Human-Machine Interface Issues

Introduction and Rationale

With the publication of the Institute of Medicine's (IOM) 2000 landmark report, To Err is Human, the public, their representatives, and the medical profession woke up to the fact that seeking medical care increases the risk of injury and death. At the time, best estimates suggested that between 44,000 and 98,000 Americans died each year from medical errors. These care-related mistakes are believed to cost the U.S. healthcare system about $2 billion each year. The prevalence of medication errors can vary greatly depending on the setting. For example, the medication error rate for hospitals was found to vary from about 0.3% overall to over 10% in a pediatric ICU setting. In addition, one estimate suggested that less than 10% of medication errors are ever reported.

One of the solutions discussed in the IOM report is the implementation of electronic safeguards in the form of computerized medical records, barcoding, and electronic medication administration records (IOM, 2000). The conversion of patient medical information into a digital format was projected not only to reduce the cost of healthcare, but to increase the opportunities for automated surveillance strategies that protect the health of patients.

To promote the adoption of electronic health records (EHR) by individual providers and hospitals, the Centers for Medicare and Medicaid Services (CMS) was given a mandate by Congress via the HITECH Act of 2009 to provide funds to help defray the costs of implementation (CMS, 2013). Eligible providers under Medicare can receive up to $44,000, while providers under state-run Medicaid programs can receive up to $63,750. Participation is not required, nor is EHR implementation, but by 2015 providers who have not implemented an EHR system would have their Medicare and Medicaid payments adjusted downward by 1% for the first year. Over subsequent years, this penalty was set to eventually reach a maximum of 5% of Medicare and Medicaid payments.

This carrot-and-stick approach would be toothless if the number of patients covered by Medicare and Medicaid were small. However, spending on Medicare, Medicaid, and the Children's Health Insurance Program (CHIP) in 2010 approached a trillion dollars and represented close to one-third of America's healthcare spending (Klees, Wolfe, and Curtis, 2012). EHR implementation on a national scale is therefore official federal government policy — one with teeth capable of chewing away at providers' profit margins if they fail to implement an EHR system and utilize it in a meaningful way.

The above policy is based on the assumption that EHR implementation will provide cost savings and improve patient safety (IOM, 2011). At the time, however, the empirical evidence to support these claims was absent. In the aftermath of the publication of several research articles revealing that implementation can increase harm to patients, the IOM formed a committee to study this issue (IOM, 2011). The committee members concluded that the patient safety benefits of EHR implementation have yet to be substantiated empirically in a consistent manner. Of the different EHR software modules that exist, the most promising for reducing medical errors were found to be computerized physician order entry (CPOE) and clinical decision support (CDS).

The IOM Committee on Patient Safety and Health Information Technology noted that adapting EHR tools to meet clinicians' needs is probably the best approach for ensuring patient safety (IOM, 2011). However, alterations in clinical workflow due to EHR implementation can impede efforts to effectively communicate patient information, increase workloads, cause alert fatigue and information overload, and precipitate EHR system avoidance behaviors, including the use of shortcuts. These problems can erode attempts to improve patient safety.

The need to better understand the information needs of clinicians has not gone unnoticed by researchers. From a theoretical perspective, there exists a clinical communications space within which clinicians share information (reviewed by Collins, Bakken, Vawdrey, Coiera, and Currie, 2011). To the extent that clinicians can communicate easily — whether verbally, by phone, or by email — a shared understanding exists that allows the concepts exchanged to be understood by the parties involved. This shared knowledge and set of skills is called common ground.

Common ground, however, is not always sufficient for high-quality care. Effective care teams are typically composed of individuals with unique knowledge and skills, but for these members to contribute meaningfully, common ground must still be established. Common ground therefore allows care team members both to communicate effectively and to make unique contributions to patient care. The overall effect is to expand the knowledge and skills of the care team and increase the quality of care. This phenomenon is called distributed cognition, and it is responsible for increasing the quality of care beyond the capabilities of a single clinician.

An EHR system could be framed as a contributing member of a clinical care team because it is capable of contributing unique knowledge and capabilities; however, the ability to make such contributions would also be limited by the extent of common ground established between the EHR system and clinicians. A priori, the magnitude of EHR/clinician common ground would be a function of both clinician training and system usability. Based on the perspective of the IOM Committee on Patient Safety and Health Information Technology, system usability is a function of implementation strategies, system adaptability by end users, point-of-care use, and usability feedback loops (IOM, 2011). However, these are not the only factors believed to influence whether an EHR system can protect or improve patient safety. The IOM Committee acknowledged that much more research needs to be done to understand how best to design, implement, and maintain EHR systems in a manner that predictably reduces the prevalence of medical errors.

This topic is relevant across disciplines, but especially so in the technology-driven critical care setting. The imposition of a poorly designed and implemented EHR system can no longer be viewed as a benign artifact of modern medicine, but as a potential threat to patient health and provider profitability that must be dealt with decisively and without delay.

Human-Machine Interface Issues and Research Questions

If it were true that converting from paper to electronic medical records improved patient safety and provided cost savings, there would be little controversy. However, according to a number of publications — including a comprehensive IOM (2011) report on this topic — there is little empirical evidence on which to base these assertions. Instead, a growing body of empirical evidence suggests that cost benefits are elusive for many and that patient safety may be at risk. A significant chasm therefore exists between past recommendations, current official government policy, and the clinical evidence being generated.

EHR systems have been predicted to provide many benefits, including increased patient safety, reduced operational costs associated with a paperless clinic, sharing of patient information among different providers, remote access to patient information in real time, and searchable databases that can be used by researchers (IOM, 2011). While these projected benefits are enticing, the most critical is patient safety. EHR systems are believed capable of reducing medical errors because handwriting becomes legible when converted into digital text, and medication orders can be transmitted instantly and legibly to pharmacists who then fill stat orders without delay. In addition, EHR systems have been designed to provide clinical decision support to help alert clinicians to risks associated with a treatment approach or medication combination.

These projected benefits are rarely realized, however. Instead, clinicians find that they become chained to terminals, communicate with their peers less, and spend less time with patients (Han et al., 2005). In addition, the workload on clinicians is often increased past the point of reasonableness because it is too intrusive and time-consuming to document patient encounters during clinic time (Grabenbauer, Skinner, and Windle, 2011). The amount of information that can accumulate in a patient's record from multiple sources can be daunting and lead to information overload. CDS alerts can be so frequent that clinicians begin to ignore them. The negative impact that EHR systems can have on clinician communications is also troubling, because in-person observations by nurses can provide invaluable insights into the treatment needs of a patient that cannot be communicated effectively in electronic form. Systems have been observed to slow during peak use periods and in some cases crash (Fernandopulle and Neil, 2010). Vendor support during such crises may be slow or absent, which can lead to seeing and treating patients "blind."

Many EHR-associated complaints concern the human-machine interface or system usability. In contrast to the expected greater legibility, complaints about character size being too small and having to use non-intuitive navigation steps are not uncommon (Tschannen, Talsma, Reinemeyer, Belt, and Schoville, 2011). The absence of standards of care adapted for EHR systems is also a problem, as nurses feel adrift in the absence of traditional cues formerly used to signal a new order from a doctor. Charting now takes place at the end of a shift or day, as nurses wait for doctors to make the necessary entries. The resulting impact on clinic workflow can sometimes be dramatic and put patients at risk for harm.

One of the more important aspects of EHR implementation is system usability from the perspective of clinicians. Usability is determined by the ease with which clinicians can navigate through patient information, how many steps it takes, and the cognitive load the task imposes (Ahmed, Chandras, Herasevich, Gajic, and Pickering, 2011). Usability has in turn been shown to be inversely associated with medical errors. Stated another way, intuitive and quick navigation to needed information reduces the cognitive load of clinicians and thus the error rate. The human-machine interface can therefore be a significant source of medical errors.

Increasing the usability of a system requires a behavioral approach that examines in detail the steps a user employs during the retrieval or entry of information. Both physical and mental actions are relevant, since the latter is proportional to the cognitive load induced by the task (Ahmed et al., 2011). Such studies have revealed that usability is a function of interface design and customizable features. In other words, an EHR system that can be user-modified to meet the needs of clinicians in a specific clinical setting while performing a specific task will impose the least cognitive load on users of the system.

As EHR vendors try to meet the various needs of clinicians, commercial systems have become more complex. This trend seems to be in direct conflict with the relationship between usability, cognitive load, and error rates described above. Clinicians who have transitioned from older, locally-designed, bare-bones systems to recent commercial EHR systems lament the simplicity of the older systems (Abramson et al., 2012). These vendors seem to be trying to provide all the "bells and whistles" that any clinician would ever need, without realizing that such efforts could be increasing the risk of harm to patients.

What seems to be needed is more research into how the clinician interfaces with the machine in specific clinical settings, in order to better understand how EHR systems should be designed. This will require detailed analysis of clinicians as they enter or retrieve information. This data could then be used to optimize EHR interfaces to reduce the cognitive load on clinicians. If EHR systems are going to make a positive contribution to patient safety and healthcare costs, then the design and implementation of such systems needs to be based on empirical evidence. Currently, such evidence is weak and inconsistent.

The research questions asked in this study are exploratory in nature, which is consistent with the relatively underdeveloped state of research on this topic. Specifically, this study is designed to document in detail the human-machine interface of a classroom EHR system as clinicians review medication error case studies. The goal is to identify weaknesses and strengths in the classroom EHR system from the perspective of experienced and well-trained nurses pursuing a graduate degree in nursing. In addition, the demographic information provided by participants will allow an analysis of cognitive load in relation to nursing and EHR experience.

The theoretical framework underpinning this study is the clinical communications space as discussed by Enrico Coiera (2000), who argues that for information to be communicated effectively and with a low likelihood of error, common ground must exist between the two parties. This common ground can consist of shared knowledge, skills, and training, similar to that existing among most clinicians. Coiera also argues that common ground must be established between a human being and a computer terminal for effective communications to take place. This implies that the person using an EHR system has received sufficient training to understand how to communicate efficiently with the software and that the information is formatted and presented in a recognizable manner. The responsibility for establishing common ground therefore rests on the shoulders of end users and the software and system designers. To conclude, the common ground established between a clinician and an EHR interface will be a somewhat dynamic process that will require periodic adjustments in the form of retraining and design modifications to ensure a safe level of usability.

Since the human-machine common ground is a somewhat rigid structure, clinicians will tend to prefer communications with other clinicians in dynamic situations when information needs may be changing in unpredictable ways (Coiera, 2000). When common ground is minimal, conversations tend to take up more time as more information must be exchanged to communicate individual bits of information. Coiera refers to this as the bandwidth of the conversation. If this principle were applied to the interactions between a clinician and an EHR terminal, then spending more time and using more keystrokes or mouse clicks to access the needed information would be an indication of a larger bandwidth due to less common ground.

The hypothesis being tested in this study is that common ground (usability) can be quantified by monitoring the human-machine interactions between clinicians as they work through medication error case studies. Since the study's participants are well versed in clinical skills, the amount of common ground shared by the participants should be large. By comparison, not all participants will share the same amount of common ground with the classroom EHR system. This variation should be quantifiable and statistically significant.

PHI Privacy and HIPAA Considerations

One of the primary considerations that should be on the mind of any clinician accessing patient information is patient health information privacy. Fernandez-Aleman and colleagues (2013) noted that for providers to realize the envisioned benefits of converting paper medical records into digital code — including lower healthcare costs, increased quality of care, research utility, and information mobility — the stored data had to be resistant to equipment failure, easily accessed, complete, and protected against unauthorized access. In the U.S., the Privacy Rule under the Health Insurance Portability and Accountability Act of 1996 (HIPAA) sets the minimum standards for protecting electronically stored patient information against unauthorized release. With respect to EHR systems in the U.S., the Certification Commission for Healthcare Information Technology (CCHIT) reviews the security protections built into an EHR system and determines whether these protections meet federal standards. If they do, the EHR system receives CCHIT certification.

The primary security questions that Fernandez-Aleman and colleagues (2013) asked when they systematically reviewed 49 research articles on EHR security were whether the systems could (1) meet the appropriate government standards (compliance), (2) allow de-identification of patient health information by researchers, and (3) encrypt the data. The authors also assessed whether users were being properly trained in privacy regulations. While most EHR systems and their users are able to make an affirmative claim to all of these criteria, the authors noted that there were still significant security concerns to be addressed in light of the vision policy makers have for protected health information (PHI). For example, accessing information across the internet makes use of the SSL security protocol available on most internet browsers. While this does provide a reasonable level of protection against unauthorized intrusions, it will not be enough to prevent intentional data mining by corporate hackers. There are also concerns about individuals storing their PHI in the cloud and whether cloud providers can realistically be prevented from accessing or copying encrypted information. The encryption of data also becomes a concern when patients or clinicians want to view imaging files on a PDA or other low-speed, narrow-bandwidth devices.

While privacy concerns are critical for all clinicians accessing patient information, a secure EHR system is still worthless if it does not meet the needs of clinicians. Probably the most important issue facing EHR design and implementation is clinical workflow and whether the EHR system should adapt to the needs of clinicians or vice versa. It will be argued in this chapter that neither extreme is optimal. Instead, clinician workflow needs should be paramount, but there is also something to be said for experimenting with the possibility that a properly running EHR system could improve workflow patterns. With this debate in mind, the following research findings provide additional elaboration on this important topic.

When Han and colleagues (2005) retrospectively examined mortality rates before and after EHR implementation at a PICU in a teaching hospital in Pittsburgh, they found that mortality was significantly increased following implementation. The system was implemented "big bang" style — that is, in less than a year — so there was relatively little preparation for the event. Next to shock, the second strongest predictor of mortality was CPOE (odds ratio: 3.71; 95% CI: 1.88–6.25). Among the factors the authors identified as contributing to increased mortality was altered admissions workflow. Prior to EHR implementation, clinicians would prepare medications and diagnostic tests in advance of the patient's arrival and critical medications were stored in convenient locations within the department. After EHR implementation, clinicians were no longer allowed to admit patients prior to arrival and all medications were stored in a central location under the control of the pharmacy. To make matters worse, computer terminals would freeze during peak periods of use. Implementation also reduced the size of care teams as one individual remained at the computer terminal, which had a negative effect on clinician communications and face time with patients.

The study period examined by Han and colleagues (2005) was limited to 13 months prior to and 5 months after EHR implementation. It may therefore be possible that mortality rates eventually returned to the pre-implementation mean as clinicians adapted to the workflow changes imposed by the CPOE system. After a team of clinicians from a PICU at a teaching hospital in Seattle traveled to Pittsburgh to learn from those mistakes, the same CPOE system was implemented — but only after careful preparation (Del Beccaro, Jeffries, Eisenberg, and Harry, 2006). In addition, the workflow needs of clinicians were placed before the needs of CPOE designers. The ability of clinicians to prepare critical medications and order diagnostic tests in advance of the patient's arrival remained intact, as did the decentralized storage of critical medications. To minimize the number of steps required to submit orders, a number of order sets were created and vetted in advance of CPOE implementation. In recognition of the negative effect EHR systems have on clinician verbal communications and patient face time, the Seattle clinicians established a standard of care stating "CPOE does not replace talking" (Del Beccaro et al., 2006, p. 294). The Seattle researchers concluded that the CPOE system itself was not the cause of the increased mortality rates in Pittsburgh, but rather how the system was implemented and managed.

Workflow Disruption and Patient Safety

In a similar study, Longhurst and colleagues (2010) examined the impact of CPOE implementation on mortality rates at Stanford University's Children's Hospital. Their reasoning was that most published studies investigating the effects of EHR implementation had revealed increased patient harm or no benefit; therefore, there was a need to test whether CPOE implementation could improve patient safety. Longhurst and colleagues revealed that a hospital-wide implementation of an EMR system with CPOE significantly reduced mortality by 20% (p = 0.03).

Longhurst and colleagues (2010) also found that when the EMR system adapted to the needs of clinicians, this fostered standardization of care standards across the organization. These care standards in turn tended to streamline workflow activities. For example, over 300 order sets were created and vetted prior to implementation, but after implementation the order sets began to impose consistency in care throughout the organization. The standardization effect of the EMR system was also believed to be responsible for more effective communications between clinicians, a finding consistent with common ground theory. Workflow took on additional dimensions as real-time patient information became remotely accessible. Time savings were realized when pharmacists no longer had to transcribe paper orders before filling them, which improved accuracy and reduced turnaround times. This benefit alone was believed to be the main factor differentiating their findings from those in Pittsburgh (Han et al., 2005). As the authors noted, experts have since deduced that the delay in filling vasoactive medication orders in the Pittsburgh PICU was the most likely cause of the increased mortality rates.

In a more recent article, Longhurst and colleagues (Hahn, Bernstein, McKenzie, King, and Longhurst, 2012) argue that a physician electronic notes system implemented rapidly across an organization had little negative impact. They reported that the system did not significantly alter physician workflow and eliminated many of the costs associated with a paper medical record system. The note system chosen was not the one recommended by their vendor, but the one that received more positive reviews in the literature. The chosen system used a one-click process that opened a note self-populated with patient vital signs, biometric measurements, and laboratory results. This system was implemented within a single year across 46 inpatient services, hence the "rapid implementation" term in their article's title.

When reading the studies by Longhurst and colleagues it is hard not to wonder whether the purpose of their publications is to promote the promised benefits of EHR implementation, such as lower costs and improved patient safety. It could also be argued that Stanford represents an elite institution and that any data generated in-house would be marginally relevant to everyday community hospitals. The findings of Brunette and colleagues (2013) seem to support this possibility and suggest that the workflow problems described by Han et al. (2005) may be the norm rather than the exception.

Brunette and colleagues (2013) examined the mortality rates for an emergency department (ED) at a large urban medical center in Minneapolis, Minnesota for one year before and after CPOE implementation. Some of the main problems encountered were an inability to place orders until the patient had physically arrived in the stabilization room, slow system speeds, reduced face time with patients, too many steps to place an order, and an inability to ensure that an order was received by a department providing a critical service. These issues, however, were not sufficient to significantly increase the mortality rate of ED patients. Still, the authors argue that no increase in mortality rates implies that the promised benefits of increased patient safety and lower cost were probably not realized within the study period. Once problems became apparent, they were addressed in various ways. In urgent situations, physicians were again allowed to place orders verbally, with such orders then retroactively placed using the CPOE system. Order sets evolved over time to become more efficient, and server speeds and capacity were upgraded. Although workflow was disrupted by CPOE implementation, system modifications were implemented to allow the needs of clinicians to again take precedence.

In a recent observational study of nurses' experience with a CPOE/CDS system in PICU and NICU settings, several workflow problems were noted (Tschannen et al., 2011). Compared with paper systems, the CPOE system was susceptible to duplicate orders. Reconciling these duplicates took additional time, as nurses had to enter notes explaining why the duplicate was deleted. The authors noted that the CPOE system itself was not the source of workflow problems, but rather how information was being formatted by the system.

In the era of paper records, nurses would notice flagged medical records or the presence of a physician on the floor, and these cues would alert nurses to check for new orders (Tschannen et al., 2011). With the CPOE system, no alert mechanism was provided. Checking for new orders at regular intervals therefore rarely occurred. In addition, the system lacked a mechanism for validating orders before medication administration. The nurses also commented that clinician communications were reduced following CPOE implementation. Overall, CPOE implementation increased the anxiety levels of nursing staff due to changes in workflow patterns.

To provide patient-centered care for patients with chronic conditions such as heart disease, diabetes, or chronic obstructive pulmonary disease, a clinical practice was created from the ground up to support the 25,000 hotel and casino workers union in Atlantic City, New Jersey (Fernandopulle and Neil, 2010). This process was begun in 2007 and was up and running by 2008. From the beginning, the clinic was designed to be paperless. Some benefits were realized immediately, including not having to search through piles of records for needed information, remote access to patient information, improved legibility, and quicker prescribing of multiple medications for the same patient. However, a year after opening, the EHR system began to slow and occasionally crash (Fernandopulle and Neil, 2010). Things became so bad that patients were being seen and treated "blind." Getting the system fixed took several weeks and intense lobbying efforts with the software manufacturer. The authors reminisced in their article that a paper system never crashed.

When the e-prescribing service was finally brought online, the clinic staff soon became aware of a large security problem in the software (Fernandopulle and Neil, 2010). Apparently, anyone could place an order with the pharmacy. It took software designers the better part of six months to fix the security problem, so prescriptions continued to be handled using paper. Although the system designers had promised the possibility of receiving laboratory results in a digital format, this was never realized. As a result, all lab reports are scanned into the system and are therefore unsearchable. Alert fatigue became a problem as even the most routine prescriptions elicited warnings. Physician workload increased significantly as physicians became the primary data-entry personnel for patient information.

Probably the most important setback realized by Fernandopulle and Neil (2010) was the medication reconciliation system. The system was designed to require a current evaluation of medications being taken by the patient, regardless of who was accessing the medical record. Unfortunately, not everyone who accesses the record is qualified or willing to be responsible for this critical task. As a result, medication lists contained missing entries and reconciliations often failed for this reason. Such problems contributed to several medication-related adverse events. In addition, the promise of data analysis capabilities was never realized, as searches were observed to retrieve incorrect results. The authors lamented that at least with the fault-ridden and cumbersome paper system of the past, any problems needing correction were within the technical expertise of office staff. With an EHR system, vendors and IT personnel are the only ones with the expertise needed to fix major problems.

These studies reveal that the promised benefits of EHR systems for individual practices may not be within reach using contemporary commercial products. Some of the promises mentioned by these authors were a paperless office, flawless e-prescribing, system-wide integration, all-digital information, searchable databases, and a level of security that exceeds HIPAA guidelines. The study by Fernandopulle and Neil (2010) revealed that workflow can be altered in unpredictable and troubling ways when such systems are implemented.

To briefly summarize, the promised reduction in costs and improved patient safety due to EHR implementation is rarely realized by everyday healthcare organizations. While many of the problems encountered in the past can be addressed, there remains a great deal of room for improvement in commercial systems. While one elite institution has reported improved patient safety and possibly reduced costs, the relevance of those findings to community hospitals is questionable. The possibility that an EHR system can be purchased off the shelf, implemented, and result in both improved patient safety and lower healthcare costs is not credible in light of the evidence presented here. Therefore, to avoid e-iatrogenesis, organizations considering EMR or EHR implementation need to heed the mistakes made by others.

These conclusions are consistent with the recommendations of Michael McBride, technology editor for the journal Medical Economics. In the August 2012 issue he states explicitly that clinical workflow will invariably be disrupted by EHR implementation. To minimize the impact, a thorough workflow analysis should be conducted before a commercial EHR system is chosen. In individual practices, this will likely require a physician to become intimately familiar with all the tasks that his or her staff performs on a daily basis. Once a detailed overview has been grasped by at least one person, minimizing workflow disruption then depends on matching the EHR system to existing workflow patterns. This recommendation implies that out-of-the-box EHR systems tend to be relatively inflexible; however, since there are a large number of systems available, choosing one that closely matches the organization should be feasible for most practices.

In a November 2012 article, McBride passed along advice from participants in a large EHR best practices study being conducted by Medical Economics. Early findings include the slowing of workflow by as much as 50% during EHR implementation. Participant comments included the following: doctors ending up spending all their time entering data and never getting around to seeing patients; doctors falling further and further behind in EHR documentation; the emergence of "EHR-only" patients (patients seen virtually only); and an unrealized promise of system integration. Two recommendations stand out above the others: (1) spend the time and money to become thoroughly trained by expert users, and (2) designate one person in the practice a "superuser."

If the findings of the studies examining the workflow needs of healthcare providers could be distilled to their essentials, they might be the following:

1. Do not make the mistake of adapting workflow to the needs of the EHR system, or patient safety will suffer.

2. Individual practices and larger healthcare organizations should perform a thorough and pragmatic assessment of workflow procedures already in place prior to EHR implementation, and then identify the product that fits best. Doing so will minimize the disruption the EHR system has on workflow and patient safety.

3. Rather than using a "big bang" implementation approach, consider implementing the EHR system in well-planned steps over a period of years.

4. Staff training is paramount and, when possible, hire the services of someone who has actual experience working with the chosen system on live patients.

5. Well-considered and thoroughly vetted order sets should help minimize the amount of time clinicians spend in front of a computer terminal.

6. Designating someone to be a "superuser" in the practice or clinical department seems to ease the transition from paper to electronic systems.