Adaptive and (Mostly) Maladaptive Responses to Computerized Health Systems
IST 780-907-200725
Org Soc Issues in Healthcare Informatics
Dr. Scot Silverstein
Linda Cook
March 2008
Abstract
The implementation of computerized health systems, including CPOEs and EHRs, change the users’ work processes, sometimes with unanticipated results. Different types of unexpected outcomes are described. A frequent consequence is disruption of communication among colleagues. Artifacts may be used as important non-electronic sources of information. Users may develop workarounds when new systems conflict with the actual work flow and task demands. Human problem-solving can resolve conflicts and plays a critical role in the utilization and effective functioning of HIT.
Overview: Consequences of Implementing New Technology
The adoption of information technology is spreading in clinical settings for benefits that include broader access to medical records, reduction in treatment errors, avoidance of redundant procedures, and improved quality of care. However, while stakeholders expect these benefits, they fail to anticipate many of the associated changes that are likely to occur with the advent of new technology [1]. These unforeseen consequences have many sources and can be categorized along several dimensions. Unintended consequences of IT have been identified by Ash (2004) [2] as related to errors in entering and/or retrieving information, and to errors in communication and coordination. Among the former are errors stemming from interfaces unsuited for interruption-prone environments and cognitive overloads caused by overly long, highly structured forms. Among the latter are loss of feedback from collegial interactions, and multiple mandatory log-ins to document inherently fluid processes such as ordering and administering medications. Koppel [3] made a similar distinction between information errors, caused by data fragmentation and lack of system integration, and human-machine interface flaws, such as nurses delaying logging medication administration to the end of their shift. Han [4] discovered serious sequelae to the implementation of a CPOE system which changed clinical communication patterns. Mortality rates rose after physicians were prevented from preparing meds based on information from ambulance teams transporting patients, but were forced to wait until the patients arrived and were registered in the system. While they devoted time to entering orders, nurses spent time activating them, taking providers away from the bedside and direct provision of care.
Ash (2007) [5] recognized that not all unexpected outcomes of CPOE are negative. They devised a comprehensive taxonomy of consequences, starting with a division of the anticipated from the unanticipated. Anticipated desirable consequences are the goals of the projects; the undesirable anticipated are the acceptable tradeoffs (cost, extra work, etc.). The unanticipated consequences may be serendipitous benefits or “unintended consequences”, a term reserved for undesirable unanticipated outcomes. Concentrating on unintended consequences as the area requiring the most attention, Ash and her colleagues found eight distinct types emerging among users of CPOEs [6]. These are 1) more/new work resulting from the system; 2) workflow issues; 3) never-ending demands for support, new equipment, etc.; 4) communication issues; 5) intense emotions aroused in users; 6) new kinds of errors spawned by CPOE use; 7) changes in the existing power structure; and 8) overdependence on technology. Weiner [7] contends that the most critical of the eight error types identified by Ash et al. [6] is the introduction of new errors to the clinical process by HIT, such as CDSSs that influence physicians to select incorrect diagnoses. He terms this “e-iatrogenesis”. Campbell [8] further explored the effects of overdependence on technology, when it is expected to do more than support clinical work. She notes that errors can arise when providers have false expectations related to the accuracy of the computerized data, or overestimate the system’s capabilities. She cites instances in which providers entered allergy data in convenient free text fields, expecting it to be available to the decision-support function. Providers may also incorrectly defer to the judgment of the CDSS: the combination of heparin and aspirin triggers alerts, but may in fact be appropriate in a coronary care unit. Caudhill-Slosberg [9] found errors propagated within EHRs when providers used cut-and-paste to copy clinical notes instead of entering fresh comments. A mishap that made the news in the UK occurred when providers trusted the NHS e-formulary‘s selection of “sildenafil” when “Zyban” was ordered - and patients were prescribed Viagra for their smoking-cessation program [10]. Nyssen [11] describes a case study in which a patient presenting repeatedly to two hospitals failed to be recognized as an emergent problem and eventually died of respiratory arrest. He ascribes this outcome in part to the use of patient records promoting a “plagiarism of diagnosis”, hindering the identification of an evolving condition by promulgating an existing problem definition.
Wentzer [12] suggests reframing Campbell’s typology, organizing errors along two dimensions based on observations of the Danish health system. She suggests evaluating implementation by analyzing the type of human-computer interaction difficulties (interruption of usability; discontinuity in utility; sociotechnical problems) across the systemic level at which it occurs (direct user-computer interface; IT-mediated interaction; IT infrastructure). The resulting matrix of nine problem types focuses attention on the range of areas affected by HIT implementation and the origins of errors.
In all the examples cited, the important factor is the way in which the novel technologies function in the context of the workplace. The interaction of HIT with people in practice determines the kind and likelihood of errors, and the ease with which errors are detected and corrected.
Disruption and Repair of Social Communication by HIT: The Use of Artifacts and Chat
When IT systems are implemented in healthcare settings, particular attention must be paid to the ways in which information is exchanged, and the impact of the technology on existing communication channels. Interactions among staff serve to coordinate patient care and cannot easily be automated. In a study of care transitions in an outpatient surgery, Schultz [13] listed the elements observed to facilitate or impede the timely transfer of patients between services. The presence of accurate, complete, and recent information – received from the EHR, paper chart, informed staff, or patient – were the most frequent facilitators, and missing, delayed, or unclear information were the most frequent barriers. Information is data that can be understood: Oleson [14] makes the point that medication ordering isn’t a simple transmission of data from doctor to nurse to pharmacist. Rather, at every point in the chain the order undergoes a translation of meaning as each agent processes the prescription according to role and task. There is a threat of communication failure if the system does not preserve the meaning across the different contexts. Hartswood (2001) [15] has shown how something as seemingly obvious as the term “mobility” has different meanings when used by an occupational therapist, physiologist, nurse, or physician. He argues that HIT should start by recognizing the many “local” meanings and deriving a “universal” meaning which is broadly understood, as opposed to beginning with a general term subject to divergent interpretations in varying contexts. Elsewhere, he describes how what a provider chooses to communicate about a patient, and which descriptors are recorded, differs depending on the purposes of the communication, such as placing a patient in a service with specific entry guidelines (2003) [16]. The assumption that the EHR is a standardized vessel of objective data is not supported.
There are circumstances in which the workplace must support communication at transition points in clinical care, despite changes wrought by the introduction of technology. The handoff of patients between shifts is an example of a situation in which efficiency of communication is vital component of the process. In a study of handover practices in a UK hospital, Kerr [17] identified social, organizational, and educational functions served in addition to informational content. Verbal handovers accommodated all these, and allowed the presenter to adapt, omitting information already known to the incoming nurses and expanding when necessary to orient less-experienced staff. When handovers were conducted by playing tape-recorded reports, time efficiency was achieved, but social and educational functions were lost. Engesmo [18] reinforced these findings, noting differences in shift handovers conducted face to face or by reading the EPR. As well as tailoring content and directing information to the responsible individuals, verbal handovers permitted the communication of items that could not be written, because of its nature or because it might be reviewed by the patient: e.g. suspected but unverified dementia, or a doctor’s non-cooperativeness. Roberts [19] describes an automated “coverage list” (developed as a tool for tracking responsibility for patients) and a computerized sign-out application for transferring responsibility at Brigham and Women’s Hospital in Boston. The system nevertheless includes face to face contact, supported by a detailed printout.
Resilient systems provide an ability to detect and correct errors at points of transitions, whether at shift change or in processing an order. Patterson (2007) [20] defines collaborative cross-checking as a strategy where members of a system with different perspectives examine another’s actions for validity and/or accuracy, such as a pharmacist verifying a physician’s unusual order. She concludes that routinized cross-checking, including software alerts, are not very helpful in detecting errors, for which specialized knowledge is frequently required. Her findings suggest that some value is lost when human communication is subrogated by automated systems. She sees an ongoing role for personnel with weakly defined roles, who can operate outside the routinized processes when necessary.
Doctors and nurses alike found their communication disrupted by the advent of a CPOE system studied by Beuscart-Zephir [21]. Before the implementation, doctors ordering medications while on rounds offered a general time for administration. Nurses in attendance probed to clarify, specify, or modify the order. When doctors initiated direct order entry, they were forced to input an exact time; nurses were unable to determine whether it was a default choice or critically precise. The nursing manager opted to double the number of laptops purchased so nurses could join doctors on rounds and maintain an active role and off-line communication, summarizing and managing their data as the doctors entered it. Koppel [3] also noted the persistence of extra-IT communication, with physicians questioning nurses directly to confirm whether medications not displayed in the system had in fact been administered per schedule. Wentzer [12] offers a case study of a semi-successful attempt to replace telephone communication with electronic messaging among hospitals, the municipality of Aalborg, Denmark, and local general practitioners as a “more quick and effective way of communication”, leading to more correct care. However, the inexperienced hospital staff found the system extremely time-consuming, and were unable to confirm that messages were delivered. Eventually it was agreed only non-urgent messages, with a required response time of more than 24 hours, would be sent electronically; all others would be phoned in, contravening the point of the exercise. A confounding factor was the personal patient consent legally required for electronic transmissions of data, which increased the work burden and raised doubts among staff concerning the legitimacy of the enterprise.
Ash et al. originally considered the persistence of paper records to be an undesirable unintended consequence of HIT implementation, but reconsidered their position [6]. Campbell [8] sees a role for paper and other artifacts in supporting efficient communication through social rather than technical means. There are several examples of artifacts serving an important informational role in parallel or in preference to screen displays. Hartswood (2001) [15] cites the practice of printing summaries of psychiatric assessments for use by departments not connected electronically, thus expanding the scope of shared information in an exchangeable format and, as he phrases it, “universalizing the local”. The sign-out application at BWH mentioned by Roberts [19] supplements the electronic patient handover with printed documentation for the incoming physician. Nurses changing from a manual transcription to a CPOE system used a printout of patients’ current medications list as an aid to prep trays for administration. The list’s informational value was enhanced by the manual strikeout of discontinued drugs (included on the list with the date of discontinuation) and by annotations documenting questions or concerns about the orders. The practice evolved to cutting off handwritten notes and using them as a supplemental order list; thus, the use of the new system matured from an imposed “rule” into a tool that supported the work practices.
Wears [22] describes an informational modality for which no satisfactory electronic substitute has been discovered. He found the status boards used in a US ER and a UK pediatrics ward shared a remarkably similar structure, being used to track the location status, and associated information on active patients. Six properties underlying their unique usefulness were extracted: the boards were 1) malleable; 2) ecological; 3) locally owned; 4) widely available; 5) informal; and 6) accessible. They provided an immediate overview of the ER/ward’s activity level. By contrast, IT attempts to display similar information were less accessible for viewing and updating, and less flexible in their use. The status boards were so vital to the department’s operation that at one site a web cam was installed to display the image in the admissions and nursing areas. Wears makes the point that IT must be cautioned against designing away useful functionality in its quest for automation. Ash et al. [2] found a similar situation in an ICU which used a large paper day sheet to post order lists, problem lists, vital sign graphs, and med lists. Replacing this unitary artifact with a CPOE that fragmented the information over several screens reportedly worked against the mental summary provided by the paper.
Using the System in the Real World: Workarounds
The gaps between designers’ intent and real-time performance of HIT only become apparent once the system has been introduced into a particular work setting. The users and their job demands will determine how good a fit exists with the application. When the system’s functionality is flawed, insufficient, or does not match the work flow, users will develop workarounds in order to accomplish their tasks by any means necessary. Koyabashi [23] defines workarounds as “informal temporary practices for handling exceptions to normal workflows”. She remarks that successful workarounds may provide institutionalized solutions to problems and become part of the accepted practice. However, unsuccessful workarounds can have a cascading effect, causing more exceptions down the line and greater deviation from normal processes. She makes reference to the success of artifacts such as status boards in coordinating work, but observes that, dynamic as they are, they preserve no institutional memory of successful practices. Wears (2007) [22] echoes this observation, but suggested this attribute might be valued by users posting information they would not want permanently stored. This is similar to Engesmo’s [18] point about verbal handoffs, allowing nurses to say what can’t be written: a use of informal but effective communication, unsuited to IT. Koyabashi further states that workarounds often develop under conditions of imperfect information. This would pertain in situations in which the CPOE and EHR record is incomplete, ambiguous, or suspected of inaccuracy. Workarounds may consist of users conducting work outside the system, or using the system otherwise than as intended. As an example of the first, Ash et al. [24] tell an anecdote in which the implementation team attempted to compel staff to use CPOE by stealing their paper order forms. The providers resourcefully used napkins to record orders. To avoid having tableware filed in the charts, the team restored the forms.