Campbell et al. (2006) provide a typology for looking just at these negative outcomes, again with a CPOE focus.1 Campbell and her team interviewed and observed clinicians (including physicians, nurses, pharmacists, and allied health care providers) at five hospital sites. Using a grounded theory approach and card sorting techniques, they first distilled shared themes from across observations and interviews. Next, in the card sort, with clinician assistance, they grouped the ideas presented into nine classes of negative unanticipated outcomes. These categories, organized as a typology, add a finer grade of classification to the discussion of UCs.
Campbell’s typology included categories for generalizing across problems, and mapped UCs to their underlying outcomes. For example, the additional work generated by technologies can vary from transformed work practices and workflow to additional required effort (e.g., documentation, the handling of new decision support alerts). When compared to the paper-based clinical practice at sites within their study, respondents noted, there is simply more work to be done. Other classes of UCs include a poor fit between the human-computer interface and the context of use, unintended overload of individual cognitive and collective work processes, and changes to coordination and communication practices. Of course, at this point in time, some of those consequences can be construed as anticipated. As with Ash et al.’s hierarchy, several instances within this typology are extensible to other forms of healthcare products beyond CPOE, including medical devices, communication tools and other technology.
While the above typologies focus on an expectation of outcomes, socio-technical models reevaluate UCs through a systems’ lens. Rather than focusing only on the technology, these models were developed with foundations in systems research, and were based on the idea that the impact of HIT can only be understood while considering its social, organizational and technical context of use (Fox 1995; Cummins and Srivastva 1977). They depict complex and interdependent components of the health care system including users’ characteristics, workflow, organizations, and policy along with the health information technology itself.
In Harrison et al.’s Interactive Socio-technical Analysis (ISTA) (Harrison et al. 2007), UCs were not seen as created by the HIT system (e.g., failure to fully understand the impact of design); instead the consequences were understood as resulting from different types of interactions. ISTA depicts the emergent relationships between HIT, clinicians and workflows. Technology is viewed as part of the complex system that is shaped by the technical and physical infrastructure within which it resides. The system as a whole is understood as the interaction and interdependence among its components. UCs in this framework are not solely classified by anticipation of their design (e.g., anticipated use/unanticipated outcomes), rather ISTA considers how HIT is actually used within a given context. Thus, interaction type is used to define UCs rather than the intent or outcome. The five interaction types include: (1) new HIT changes existing social system, (2) technical and physical infrastructure mediates HIT use, (3) social system mediates HIT use, (4) HIT-in-use changes social systems, and (5) HIT-social system interactions engender HIT redesign. Instances of new HIT changing the existing social system include UCs such as new/more work on tasks such as documentation, changes to informal interactions yielding communication changes, or alterations in workflow such as shifts in roles and responsibilities. As illustrated in Table 11.1, Harrison et al. incorporate both Campbell’s typology (Campbell et al. 2006) as well as the work on communication and information transfer by Ash et al. (2007, 2009) into their interaction types. Importantly, ISTA shifts the focus from causation or outcome of UCs to pointing out the impact and differences of systems in use from the ways in which the systems were designed. Harrison’s framework offers a richer and more nuanced analysis, and provides significant potential for remediation through redesign.
Table 11.1
Unintended consequences by ISTA type
ISTA type | Unintended consequencesa |
|---|---|
1. New HIT changes social system | More/new work for clinicians b |
Physicians spend more time on documentation and justification | |
Changes in communication patterns and practices | |
Introduction of IT leads to decline of vital interactions among care providers, ancillary services and unitsc | |
IT system eliminates informal interactions and redundant checks that help catch errorsc | |
Workflow | |
CPOE undermines informal gatekeeping by clerk who decided whether patients really needed daily x-rays | |
2. Technical and physical infrastructure mediate HIT use | Paper persistance b |
Paper used to solve problems of lack of integration of CPOE and other clinical information systems | |
3. Social system mediates HIT use | New types of errors b |
Busy physicians enter CPOE data in miscellaneous section rather than scrolling for optimal location. Improper placement can impede use by other physicians and by CPOE systems | |
Causing Cognitive Overload by Overemphasizing Structured and “Complete” Information Entry or Retrievald | |
Fragmentation | |
Distribution of information over several screens sometimes leads busy physicians to miss key parts of record, such as interpretations or reports by other types of physicians | |
Structure, overcompleteness | |
Extensive reporting requirements lead physicians to cut and paste whole reports, rather than extracting pertinent facts | |
Paper persistence b | |
Counter to hospital directives and recommended IT practice, MDs who prefer paper records annotate CPOE printouts and place these in patient charts as formal documentation | |
Misrepresenting collective, interactive work as linear, clear cut, predictable workflow d | |
Inflexibility: Transfers: Inflexible EHR reporting requirements generate failures to record clinically appropriate drug administration and cause difficulties in managing patient transfers | |
Urgency: Nurses and Physicians refuse to follow data-entry rules requiring physician pre-authorization for urgent care | |
Workarounds: Physicians and nurses provide urgent care by working around cumbersome procedures | |
Misrepresenting communication as information transfer d | |
Decision support overload: Alert fatigue: physicians ignore warnings and reminders | |
Loss of communication: Urgent requests and some test results from accident and emergency, admissions are never viewed on ward terminal | |
Loss of feedback: Nurses initial orders on receipt, rather than administration, so physicians cannot tell if orders have been carried out | |
Human-computer interface unsuitable for highly interruptive context d | |
Juxtaposition errors | |
Entry of orders for or on behalf of the wrong person | |
4. HIT-in-use changes social system | Changes in the power structure b |
Narrow, role-based authorizations redistribute work – requiring physicians to enter orders directly | |
Remote monitoring by the organizations undermines physicians’ autonomy | |
IT, quality assurance departments, administration gain power by requiring physician to comply with CPOE-based directives | |
In decentralized systems, internal variations in CPOE uses and configurations increase interdepartmental conflicts and competition | |
5. HIT-social system interactions engender HIT redesign | Never-ending system demands b |
As implemented CPOE systems evolve, users rely more on the software, demand more sophisticated functionality, & customize software (e.g., physicians create their own order sets). New features must be added to original software. Interactions among multiple variations of the software in use make CPOE system unmanageable & require replacement with newer versions |
The 2009 American Medical Informatics Association (AMIA) Annual Health Policy meeting focused on outlining “outcomes of actions that are not originally intended in a particular situation (e.g., HIT implementation).” The resulting publication (Bloomrosen et al. 2011) from a panel of experts considered another perspective on sociotechnical systems and consequences. In their article, Bloomrosen et al. put forth a model with inputs and outputs that span domains including:
Technology: hardware and software systems that are implemented and the constraints they impose.
Human factors and cognition: thought processes, habits of behavior, and mental capabilities that humans bring to the use of HIT tools and processes.
Organization: embedding of technology in the complex environment of healthcare organizations.
Fiscal/policy and regulation: the legislative and regulatory environment governing the design, implementation, and use of HIT such as HIPPA requirements, indicators of meaningful use and standards for health information exchange.
In this input-output model, interactions define the model as they did in ISTA. The domains of technology, organization and human factors along with the addition of policy and regulations converge into a sociotechnical system with an even broader scope and in which HIT resides. Complicated interactions yield outcomes that can be understood in terms of types of consequences and the affected stakeholders. Like the ISTA framework, Bloomrosen’s efforts frame UCs as a study of interactions. The input-output model specifies stakeholders (i.e., inputs) as well as results or outcomes as components within the sociotechnical system. The complexity of the system underscores the need to understand points of input to the unintended consequences. For example, poor usability of an interface can increase the cognitive burden on the clinicians by requiring searching for a returned laboratory value in a sea of electronic, scanned, and paper data. Cognitive factors such as limited memory and attention coupled with a poorly designed or cluttered interface may engender potential UCs. These inputs can lead to output (i.e., consequences) that may impact both cognitive (e.g., diagnostic reasoning) and care processes for patients and providers. At another level of analysis, organization policy may serve to mitigate or exacerbate these consequences. In this example, documentation requirements could lead to workflow changes generating further unanticipated outcomes.
This multi-faceted model depicted in Fig. 11.2 underscores the shifting view of UCs as an individual problem to a perspective in which UCs is considered as complex and situated in system-wide issues. Embedding HIT into sociotechnical frameworks highlights the need to consider all the interactions of inputs and products of work in design.
11.3 Exploring Unanticipated Consequences
The potential unintended impacts of HIT in clinical settings are wide ranging, including risk of harm to patients and inefficiencies in work practices. Just as UCs can occur with technology (Tenner 1997), introducing new devices, or new processes, have the potential for both gainful and harmful effects beyond the expectation of the product developers. CPOE-based problems are some of the best documented, and are some of the most documented issues related to unintended consequences of the use of healthcare technology (Ash et al. 2003, 2007, 2009; Campbell et al. 2006, 2007; Koppel et al. 2005; Weiner 2007).
Some of the technology-induced errors are derived from the user interface. Reckmann and colleagues’ (2009) review identified problems created by poor usability including incorrect drug selection induced by lengthy drop down menus (Shulman et al. 2005), and duplicate orders or failures to discontinue medications (Koppel et al. 2008). Subsequent problems also arose when unexpected consequences led to downstream issues. For example, Computers on Wheels (CoW) were used to seamlessly move computers to the patient’s bedside. Having a computer at the point of care can potentially prevent errors in identification, reduce interruptions, and improve the completeness of procedures such as documentation. Combined with bar code technology, CoW can improve medication administration by reducing medication errors (i.e., scan the patient, scan the medication to prevent errors). However, Koppel et al. identified 15 kinds of modified workflows in use while using the barcode medication administration technology (19). For example, the authors identified an instance where the potential benefits of bar code/CoW systems were thwarted when these units were too large to fit into the patients’ rooms. Rather than scanning patient wrist identification at the bedside, nurses would print out extra bar codes outside the room. Such alterations to clinical practice can have downstream effects and, in fact, can lead to identification errors this technology was originally intended to prevent.
Similarly, HIT may not function as expected in the realm of Clinical Decision Support (CDS) alerts. Alerts for drug-drug or drug-allergy problems can be triggered during the prescribing process. If too many alerts are delivered, clinicians may fail to acknowledge the appropriate and relevant alerts. Additionally, the high rate of potential notifications can lead to alert fatigue (Steele and DeBrow 2008), and subsequently to technology-induced errors. For example, many studies found that drug-drug and drug-allergy alerts were often overridden. Payne et al. (2002) found an 88 % override rate for drug interaction alerts, and a 69 % override rate for drug-allergy alerts. Similarly, Weingart et al. (2003) found ambulatory physicians overrode 91 % of drug-allergy alerts, and 89 % of high-severity drug-drug interaction alerts. A percentage of these alerts may have provided limited detail (e.g., notifying the physician that no drug information was available in system) and perceived to be uninformative, presented information with unknown clinical significance (e.g., lacking indicators of the severity of an interaction), or may have repeated the content of previous messages.
In a 2013 case study, Carspecken et al. (2013) described an instance where a 2-year-old child was admitted to a pediatric intensive care unit (PICU) with a documented antibiotic allergy. Over a 1-month period, more than 100 alerts related to a drug-allergy cross reactivity were overridden, as the treatment was deemed as a requirement for the patient’s condition (i.e., ignoring what was considered an inappropriate alert). Over time it was determined the child did in fact have an allergic hypersensitivity, and his medical record was eventually amended. However, even after this change, the now appropriate alert (i.e., acknowledging that the child does have an allergy) was still overridden. Due to the routine rejection of the alert, clinical staff had become de-sensitized to drug-allergy alerts in this child’s case.2
There are multiple issues at the heart of this example. First, the unintended consequence of new/additional work led to an increased burden on the physicians. Subsequently, the repeated alerts decreased clinicians’ sensitivity to the message resulting in inappropriate persistence of behavior (i.e., continued override of the alert.) Additionally, the EHR did not make the addition/change to the allergy list salient to the users of the system. Finally, the unintended changes to the workflow, particularly around communication practices regarding medications, may have led to less feedback and decreased opportunities to prevent this error.
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
