The impact of electronic health records on diagnosis

Access to care

One of the most dramatic changes enabled by health IT and the EHR is that the site of care can move out of the physician’s office and into the patient’s home. By taking advantage of electronic portals [7] and asynchronous communication, patients can communicate with their health care team; new problems can be diagnosed as they arise, potentially avoiding hospitalization or readmission. Many envision the day when the scheduled “office visit” is a relic, replaced by just-in-time services provided remotely, using health IT and EHR as the vehicle for providing and documenting care. Telehealth functionality (see discussion below in “Collaboration for Diagnosis”) offers the opportunity to provide real-time consultation with experts in an ever-expanding list of subspecialties, another important demonstration of health IT improving access and quality of care.

Information gathering and access

Paper records are hard to read, easy to lose and seldom complete. Electronic health records, in contrast, potentially provide immediate, reliable and readable access to the patient’s health history. Besides these very welcome improvements, the EHR has enabled new functionalities that could not have been imagined with paper records. EHRs enable new care models, such as team-based care, where different members of a care team can contribute to clinical documentation from different sites and at different times [8]. The EHR allows team members, dispersed in location and time, to see the same information, communicate asynchronously and share in creating care plans, functionality that would have been far more challenging, if not impossible, with paper-based records. As other examples of the novel conveniences the EHR enables, images can be attached to notes to illustrate a finding, and records can be electronically searched to find a specific date or provider or diagnostic test.

Many different applications are emerging to improve data collection for the electronic record. Automated tools that assist in obtaining the key elements of the patient’s history, such as “smart” clinical documentation forms in EHRs, have proven to be useful, and they complement the information gathered through in-person clinician-led interviews [9]. A further advance is to make it easier for patients to provide these data using kiosks and tablets in the waiting room. Practices are also experimenting with ways to interface with various types of patient-generated personal health records [10]. Health information exchange resources will improve access to remote data, expediting diagnosis and reducing unnecessary retesting [11].

The organization and display of information

EHRs have the potential to improve diagnosis by making it easier to organize and display data in more meaningful ways. Note templates can be created that allow a specialty to customize their documentation for their own purposes. A particularly welcome feature that has been shown to improve information processing is the ability to display test data as a graph, instead of as a text result or a text list. Graphic displays facilitate identification of more subtle trends and patterns [12].

Decision support

Decision support, at least in a rudimentary form, was available with paper record systems. A sticker might be taped to the front of each chart as a reminder that the patient needs a flu shot, for example. Well-developed electronic record systems allow for implementing that functionality at a new level of sophistication and reliability, providing reminders for a host of preventive services, or problems that need scheduled follow-up, or providing for prominent display of information relevant to a decision the provider is about to make.

Many of the earliest health IT products focused directly on diagnosis [13], [14], [15], [16], [17], and others focused on treatment and management problems, such as antibiotic selection [18], and ventilator management [19]. The most prominent examples of decision support products that support improved diagnosis are the “symptom checkers” that assist in generating a differential diagnosis. These products provide suggestions on reasonable diagnostic possibilities to consider based on the key findings in a case, and these suggestions are available within seconds. Evaluations of these products have produced mixed results [20], [21], but at least some of these products improve the accuracy of diagnosis [22], [23], [24], and improve patient care [25]. Unfortunately, even when well-designed products are available, they remain underutilized [26], [27].

Another rapidly-expanding field is computer-aided diagnosis, and these software algorithms are being used in a wide range of specialties, including dermatology, and medical imaging for breast, colon and lung cancer [28]. Using “deep learning” approaches, computer-aided diagnosis improves both detection and characterization of visual abnormalities, such as lung or breast nodules, or pigmented skin lesions, and assigning these a diagnosis.

Tools and calculators to assist in clinical decision making

Clinical decisions can involve calculations that embedded tools embedded in the EHR can simplify and facilitate. Does a patient have renal insufficiency? The EHR can immediately calculate an estimated glomerular filtration rate knowing the patient’s age, sex, weight, and measured creatinine [29]. Does a patient have a high risk for atherosclerotic cardiovascular disease? The EHR can provide a calculated risk score using the Framingham formulas for this purpose. Hundreds of other examples can be found, for example, calculation of a risk score based on probability of a specific disease, or a clinical guideline to decide whether a patient should be further evaluated, treated as an outpatient, or requires hospitalization [30].

Intelligent selection of a testing strategy

There are currently over 4000 selectable laboratory tests, and a comparably bewildering number of imaging options. Although clinicians are highly competent in evaluating common complaints and conditions, for unusual conditions or ones where a choice of investigative options exist, electronic record systems can help simplify and guide these selections. EHRs can access and integrate, for example, recommendations on laboratory test selection from reference laboratories [31]. Similarly, preferred imaging strategies are available from online resources provided by the American College of Radiology [32]. Many healthcare organizations now require that clinicians use these electronic utilization management programs to improve appropriate test selection [33].

Reliable follow-up

Diagnosing a patient’s condition can take seconds, hours or months. EHRs facilitate reliable follow-up through reminder systems that track which patients need to be seen again and when. In some advanced EHRs, clinicians can create their own reminder list of patients and tests to follow-up on at a later date. Advanced EHRs also facilitate the creation of internal registries with the same purpose, for example, a list of all the patients who need a follow-up colonoscopy at a later date either for screening or follow-up of a previously abnormal study.

Collaboration for diagnosis

Consultation from specialists is an integral part of many diagnostic evaluations. EHRs and health information exchange make it easier to share information back and forth between the referring and consulting physicians. Besides expediting the request for consultation, a shared EHR allows consultants and collaborators to see beyond the consult request per se and review whatever information may be relevant and necessary. Supplementing the traditional phone call, some organizations provide direct communication in real time using secure messaging systems, or secure directed data transfer using standards such as the direct edge protocols [34].

Telehealth

Telehealth offers another approach to improve collaboration for diagnosis and to improve the quality of health care in general [35]. Timely expert consultation is a requirement for reliable diagnosis. Delays and difficulty accessing specialists have been cited as factors contributing to diagnostic error [36], problems that telehealth and electronic records can help to solve. One of the most successful applications of telehealth is to facilitate access to retinal specialists to diagnose diabetic retinopathy [37]. Teleradiology is another dramatic success story for diagnosis, providing access to appropriate specialists around the clock [38]. For teledermatology, “store and forward” remote consultation has been found to result in more timely diagnosis than the more conventional referral process, although remote consultation was inferior to “in person” examination for accurate diagnosis for pigmented skin lesion [39].

Measuring diagnostic performance and providing feedback

Electronic records make it easier to track populations of patients with the goal of monitoring diagnostic performance and safety. How many patients in our organization received appropriate follow-up of abnormal imaging? How many of my diabetic patients have not had a timely evaluation for retinopathy? Electronic records allow analysis and reporting at the level of a given physician, team, practice site or across an entire organization, functionality that provides the basis for modern organizational quality management. The SureNet program implemented in the Kaiser-Permanente Southern California system is an elegant example of how population-level monitoring of data in the EHR can detect patients at risk for harm for diagnostic error in time to prevent it [40].

Inaccurate, inadequate and missing information

In the days of paper records, the dominant problem was simply finding it – the patients chart was too often “somewhere else”. The EHR has helped solve that problem, but imperfectly. An example is the persisting problem in communicating critical test results: Appropriate follow-up does not reach 100% even in organizations with sophisticated and mature EHRs [52]. In one study, over 10% of critical alerts were never acknowledged within 30 days [53]. Similarly, a systematic review of laboratory tests pending at discharge from inpatient care found that 20%–69% lacked evidence of follow-up [54].

Information exchange to and from physicians’ office can be disrupted by problems with Internet connectivity, system security and firewall restrictions, and granting access to unlicensed personnel to access the data. Ambulatory practice sites may be only loosely associated with the organization (e.g. hospitals) responsible for legislating their practice and system functionality.

The “copy-paste” problem

The “copy-paste” problem is one of the most ubiquitous and the most troubling concerns in using EHRs [55], [56], [57]. Electronic documentation makes it easy to reuse content from an earlier note. If modified and updated appropriately, this functionality can be a time saver for busy clinicians. Too often though, copy-paste functionality is used carelessly to summarize parts of the existing medical record, or in place of performing one’s own patient history or physical examination. Many organizations and authors have offered advice on minimizing the copy-paste problem, including comprehensive white papers recently sponsored by NIST [58], AHIMA [59] and the ECRI Health IT Partnership [60].

The copy-paste mentality degrades diagnosis in other ways as well. To the extent that documentation is inaccurate or outdated, the credibility of the entire electronic record is called into question, compromising the trust that underlies medical professionalism. “Note bloat” is a related but separate problem that impairs diagnosis by requiring the clinician to find the nuggets of key data buried in pages of unnecessary documentation. Finally, the too-facile recycling of old information tends to inhibit the ongoing questioning and reascertainment process that helps monitor diagnostic accuracy as illnesses evolve over time.

In stark contrast to the problems encountered with paper records, where charts were often incomplete or missing, EHRs create the opposite problem: information and cognitive overload. Note bloat and copy-paste problems contribute to this, along with an avalanche of alerts that providers need to digest and triage. Along with the myriad distractions and interruptions that characterize modern medicine, these factors contribute to cognitive overload, one of the major threats to reliable diagnosis.

Usability problems

Usability problems refer to design choices that hinder efficient and effective work flow, and the current generation of electronic record systems is replete with problems of this type:

Pick errors: represent situations where the wrong selection is made from a long list on a drop-down menu. Similar errors were possible in the era of paper records as well (e.g. choosing the wrong chart) but have proliferated in the electronic era. It is just too easy to pick the wrong medicine from a long list, the wrong test or even the wrong patient [61]. Solutions are starting to emerge that address pick errors, for example, EHR interface terminology with improved search terms for test selection, Tall Man lettering for medication choices or using the patient’s picture as a way to ensure the correct record was selected [62], but these innovations have yet to achieve broad adoption.

Design features that discourage reflection and thoughtful care: One goal of electronic record systems is to facilitate documentation that supports and optimizes billing. This can lead to design choices in the EHR that work at cross purposes with the clinical goal of promoting thoughtful, reflective care [63], [64]. An example is the requirement in many EHR systems to identify a specific diagnosis, or symptom, for each care encounter, and to choose this designation from a discrete list of ICD codes. This approach is ideal for electronic encounter capture for billing purposes but inhibits the clinician’s ability to consider a differential diagnosis, to explain why a certain diagnosis or testing strategy was selected or to document his or her thinking about the case. These thoughts require a narrative; they cannot be captured adequately in structured data. In addition, this too-early designation of a code for diagnosis becomes a diagnostic label that is likely to stick with the patient whether it is correct or not, and that discourages further consideration by subsequent providers the patient may see [65], [66]. We need to disentangle the roles EHRs have for billing and the roles they have for clinical care.

The EHR takes too much time: The time spent attending to the documentation requirements in the EHR adds up [67], to the point that physicians are spending as much time on the computer as with the patient [68], [69]. Replacing the information gaps that were so common with paper charts, there is now too much information, too many documentation requirements and too many alerts [70], all of which require time. A recent survey of 2509 primary care physicians found that providers received an average of 63 alerts each day; 69% of the respondents felt that the number of alerts was excessive, and 70% reported receiving more alerts than they could effectively manage [71]. Many physicians have lamented that if there were one thing that would improve diagnosis the most, it would be having more time. If the EHR is consuming time, instead of saving it, this is not good for diagnosis.

The impact on interpersonal communication and relationships

One of the fundamental concerns with the electronic record is that it can distance clinicians from patients, or from each other, altering the human element that is critical to effective care [64], [72]. This is most obvious in the examination room, as the clinician simultaneously tries to use the EHR while also interacting with the patient. The personal connection to the patient suffers, starting with loss of eye contact. Unless the physician makes special efforts to combat the problem, the patient rightly feels neglected. The problem persists long after the initial encounter, especially for inpatients, where clinicians may spend an inordinate amount of time taking care of the “i-Patient,” at the expense of time spent with the real patient [73], [74]. The electronic record, inadvertently, has effectively and perhaps inexorably compromised the patient-to-physician relationship, the foundation of effective health care.

The EHR has also created walls, electronic silos [75], that distance clinicians from each other. Before electronic records, it was commonplace for clinicians to visit the clinical laboratory, the radiology department or a consultant to discuss results from laboratory tests or imaging, or a referred patient. These rich discussions, and the personal relationships that were established as a result, have essentially disappeared as the EHR has become the de facto vehicle for communicating information [72].

The true impact of this cultural change has yet to be evaluated, but it seems likely that disrupting face-to-face communication will have negative consequences. The result will be episodes like the one illustrated by the first patient misdiagnosed with Ebola in the United States. The ER triage nurse who first interviewed the patient succeeded in obtaining a history of recent travel to an endemic region. This key finding was duly recorded in the EHR, an entry that was missed by the treating physician, who sent the patient home, delaying the diagnosis of his infection [76]. In the pre-EHR era, there is a good chance the nurse would have mentioned this key bit of data to the physician personally, or it would have been front-and-center on the paper ER note.

Ideally, the medical profession would prescribe the ideal culture to optimize the quality and safety of clinical care, and health IT products would be designed to support this. Instead, we too often are required to compromise our cultural norms, or somehow adapt them, to accommodate suboptimal health IT systems. In the Ebola case, for example, the EHR would hopefully support bidirectional nurse-to-physician communication, not inhibit it.

Other issues: Space precludes us from discussing a host of other issues that detract from diagnostic quality, including inconsistent and incomplete vocabularies for structured data elements, and variability and lack of specificity for how diseases are defined and designated. EHRs require constant maintenance. Decision support rules need to constantly updated, as an example, and failures can lead directly to errors and harm [77]. A final problem is the lack of a “gold standard” for evaluating health IT and EHR functionality. Evaluations of decision support products to assist with differential diagnosis, for example, are currently limited to comparing different systems to each other, not to a gold standard [78], [79].

Better support of clinical reasoning

Better support of clinical reasoning would be a welcome and valuable feature in future record systems. This should begin with moving away from documentation based on menu-based selections, relying more on the richness of free text notes. Capturing the clinician’s clinical reasoning is an essential element in support of diagnostic quality. “Downstream” clinicians will have a better understanding of these earlier considerations and will have a better concept of which diagnoses are truly established and strongly supported, which are tentative and need further consideration. Clinical documentation should allow the clinician, and others reading a note, to acquire a sense of the certainties and uncertainties that existed at a particular point in time, and how these were considered. Advances in natural language processing (NLP) can bridge the gap between free text notes and the structured data needed for administrative purposes and quality management. Use of NLP could allow EHRs to retain free text narratives and the rich dialogues that capture the clinical reasoning that underlies diagnosis, while still being able to extract out the structured-data elements that are needed in support of billing and quality monitoring.

Decision support should be intelligent and able to provide context-relevant information “on the fly”, decreasing the need for ad hoc searches of the medical record [82]. Ideally, the EHR would know and anticipate the needs of the user and present the right information, organized in the right format, at the right time, to optimize clinical workflow.

Improved problem lists

Improved problem lists would also be welcome [83], [84], [85]. EHR problem lists tend to resemble their paper counterpart in being outdated and inaccurate. Ideally, assuming providers could agree on the key principles, problem lists in the future could be automatically updated, could convey some sense of the support for given diagnosis and its probability of being correct and would be cleansed of any previous diagnoses that are inactive.

Patient engagement

Medical record systems will facilitate patient’s accessing and contributing to their own medical record. There is growing evidence that engaged patients have improved care outcomes [86], and more reliable diagnosis could be one of these. Engaged patients could help populate the background information on their medical history, preventing the documentation errors that now exist with documentation that is “second hand”; it is likely that a patient’s own documentation will be more accurate than what the clinician tries to capture and record. The “open notes” initiative is expanding rapidly, providing patients with “read” functionality, and the ability to now act as their own safety net to make sure documentation is accurate and that important diagnostic test results are not overlooked [87]. Patient-facing applications have enormous potential [88]; “open notes” is just the beginning.

Predictive analytics

The ability to integrate and learn from “big data” would enable several novel applications that could improve diagnosis. One possible application is to “push” reasonable diagnostic possibilities given the key findings of a patient with a new concern, for example, combining data to enhance early detection of sepsis [80]. Conversely, one can envision “error checking”, where proposed diagnoses that are in conflict with data in the EHR could be identified. Second, “big data” applications could enable population-level comparisons, where all of the patients with a given diagnosis could be compared to identify outliers that merit further investigation, or common trends that would never have been obvious in single-patient observations.

Using health IT to prevent diagnostic error or harm

Electronic data create an opportunity to identify patients at elevated risk for diagnostic error and either prevent the error or mitigate harm. The SureNet (formerly the Safety Net) program at Kaiser Permanente Southern California, for example, identifies patients with red flag findings suggestive of colon cancer (positive tests for fecal occult blood, or new iron deficiency anemia) who have not had appropriate imaging or endoscopy to investigate the problem and alerts providers to make sure the findings are followed up [40]. Similarly, “trigger tools” that search out electronic data can follow the track of abnormal laboratory test results sent to providers to ensure the abnormalities are acknowledged and acted upon [89]. The value of trigger tools to improve diagnosis has been confirmed in prospective randomized trials, effectively reducing the time to evaluation for patients with prostate and colon cancer [90], [91].

Promoting expertise and combatting overconfidence through feedback

Developing expertise requires accurate and timely feedback on performance, and diagnostic performance would similarly benefit from improved feedback [92]. Feedback is also the best antidote to clinician overconfidence that develops in the absence of meaningful feedback [93]. In the absence of any feedback, physicians, like everyone, assume that their answers are all correct [94]. Medical record systems in the future could facilitate feedback by informing physicians that their patient’s diagnosis has changed, for example, as the patient moves from an ambulatory clinic to the ER and then the ward or the ICU. This type of automated systematic feedback on diagnostic performance has great potential but is virtually non-existent [6], [92]. El-Kareh et al. [6] found only one study reporting the impact of systematic feedback on clinical diagnostic performance [95].

Solving the “i-Patient” problem

A large challenge for the staff who design and use electronic records is to somehow restore the human interaction essential to the patient-clinician relationship. Thoughtful ways to incorporate the EHR in patient-focused practice may be part of the answer. One simple suggestion is to place the display screen so that both the physician and the patient can see it and interact with it equally [96], [97]. Another promising approach is to use scribes, who can interact with the EHR, freeing the clinician to focus on the patient [98]. These are just first steps, and just in the ambulatory setting. EHR designers should strive to design systems that allow us to spend more meaningful time with patients in all settings, not less.