Healthcare is an information rich environment . It is complex, involving diverse, interdependent, knowledge-intensive disciplines, and is dynamic, involving knowledge that is being constantly revised and extended [2, 3]. This information environment extends beyond clinical care to include management, funding, and policy [1, 4] and beyond the hospital to include home care and primary care .
Policymakers have promoted the development and use of computerized information systems under the assumption that they will improve quality, efficiency, and safety of healthcare [5, 6]. Indeed, at first glance, computerization of such an information environment would seem to offer enormous organizational, safety, performance, and economic benefits. There is, however, a troubling discrepancy between the attitudes of different professional groups to healthcare information systems, with clinicians being less satisfied than information technology professionals  and less satisfied than health plan executives . Notably, healthcare professionals are often positive about the introduction of innovative technology, only to become disenchanted when they discover that it disrupts their workflow .
The dissatisfaction of clinicians with healthcare information systems suggests that many efforts at computerizing healthcare information have not fulfilled their potential, at least in the realm of patient care. Greenhalgh et al.  found no evidence that computerized information systems produce the anticipated healthcare benefits, while Greenhalgh and Stones  question the assumption of policymakers that such systems will improve quality, efficiency and safety. Furthermore, clinicians have experienced them as disruptive and inefficient [9, 10]. More recently, usage rates have been reported at above 90%, but dissatisfaction with the impact of healthcare information systems on workflow and patient throughput remains high .
The current approach to development and design of healthcare information systems is guided by a rational, technocratic world view [2, 4] that substitutes designer judgment for clinician judgment . The design solutions that emerge from this rational model rely heavily on overly-simplified decision-support rules that disrupt clinical workflows, and on inflexible templates and standardized protocols that do not accommodate the complex and dynamic challenges posed by diverse health issues [4, 10]. The ensuing systems fail to account for the demands of collaborative clinical work in which healthcare professionals must contextualize and prioritize knowledge to cope with multiple workflow possibilities and non-routine conditions . This gulf between the reality of clinical work and how it is rationalized for information-technology design leads to development of systems that induce error and do not properly support the work as intended [2, 4, 10].
As reported by Challenger and her colleagues, the experience of the UK National Health Service in implementing its Care Records Service is illustrative . The Care Records Service is an electronic healthcare record developed to manage medical records for all patients in the system. The overarching vision for the Care Records Service evolved from an unquestioning belief that healthcare needs to take advantage of new technological developments . Better patient care would emerge naturally from deployment of automation technologies that would perform specific healthcare tasks and from a suite of information technologies that would allow healthcare practitioners ready access to the most up-to-date patient information.
Policymakers assumed that transfer of tasking to automation and ready access to current patient information would improve clinical decisions via more efficient consultations, and thereby promote more effective and more economical care while reducing the risk of medication errors . Nevertheless, despite the many apparent benefits of computerization, there was widespread evidence of issues relating to the cognitive work; a lack of compatibility of system configurations with clinical practice, incomplete and inaccurate information, a restrictive data entry strategy, and an electronic-notes function that increased the cognitive work associated with taking a patient history, to name just a few . There was little evidence of the anticipated benefits.
In large part, the responsibility for the design, development, implementation and modification of healthcare information systems is left to software engineers and information technologists whose appreciation of the scope and complexity of healthcare work-as-done is necessarily limited [4, 10]. Experienced administrators and clinicians who select from commercially available systems typically have little to no expertise in device evaluation . Opportunities for healthcare professionals to influence the design are fragmented at best and typically amount to little more than an incomplete list of functional requirements as developed by a small, non-representative, albeit knowledgeable group of healthcare specialists . Nothing about this process approximates a design strategy that could lead to a coherent, robust and effective information system to support the diverse and complex demands of healthcare work.
Although healthcare information systems are focused on the cognitive dimensions of the work , they reveal no sensitivity to an insight from the field of situated cognition; that workers converge naturally on robust and powerful ways of doing work that differ markedly from the formal strategies that emerge from the technocratic world view [14, 15]. Nor do they reveal any sensitivity to the insight from Naturalistic Decision Making that experts in the field, who are making critical decisions under high workload and time pressure, do not follow a rational strategy of options analysis but rather, recognize and act spontaneously, choosing a different strategy only if the first proves to be unsatisfactory .
Absent sensitivity to these insights, healthcare information systems will, at best, fail to support the informal but powerful strategies used in patient care. At worst, healthcare information systems will disrupt and block these strategies, thereby inducing new and unanticipated systems errors and forcing those involved in patient care into work patterns and work arounds that are fragile, error inducing, and labor intensive.
There has been a recent call for healthcare to embrace the principles of high-reliability organizing [17, 18]. Despite undertaking complex and risky work, high-reliability organizations achieve exemplary levels of safety . They do so in part by remaining mindful of the subtle and complex details of operational work [20, 21]. In that respect, procedural constraints embedded in technology should not interfere with operational work and institutional demands should not be prioritized over operational demands . Technology is a crucial element of today’s high-reliability organizations , but those who design computerized information systems for healthcare impose constraints on how work is done with meager understanding of the operational work at anything more than a superficial level and with little sensitivity to its operational demands. Unless that changes, the push for high-reliability organizing in healthcare will never be completely satisfied.
User-Centered Design is offered by many as an intervention that will lead to the development of systems that will provide better clinical support . It is questionable, however, whether a well-developed User-Centered Design process involves substantive design activity, where the term design refers specifically to those activities used to formulate the design solution as distinct from activities of analysis and assessment. Ellsworth, Dziadzko, O’Horo, et al. , although critical of the tendency within User-Centered Design to enter the development cycle late, nevertheless view User-Centered Design as directed at assessment of already-designed systems.
A systematic and comprehensive process of design proceeds through the stages of defining the problem, analyzing the work, generating solutions, and assessing prototypes and fielded systems; a nonlinear process that involves considerable iteration over adjacent and nonadjacent stages. LeRouge and Wickramasinghe  identified six stages for User-Centered Design (planning and feasibility, requirements, design, implementation, test and measure, and post-release) that correspond approximately to this process. Of fourteen activities identified for the design stage, only two suggest any form of design activity, the remainder being either analytic or assessment activities. Possibly, because of healthcare’s reliance on this limited view of design, clumsy and labor-intensive features such as data entry windows, drop-down menus, lists with check boxes, and alarms dominate as design features.
Decision-Centered Design  emerged from research into Naturalistic Decision Making, the study of how people make decisions under time pressure and uncertainty . The focus in Naturalistic Decision Making is on operational work that has meaningful consequences as undertaken by experienced (often expert) practitioners. Notably, decisions made in this context may have ambiguous or poorly-defined goals. Naturalistic deciding came to be viewed as a macro-cognitive process and the insights generated through its study prompted an extension of the naturalistic method of investigation to other macro-cognitive processes .
According to Crandall et al. , the term macrocognition refers to the collection of cognitive processes and functions that characterize how people think in natural settings. The designation as macro signifies that these cognitive processes relate directly to work goals. Cognitive processes such as situation assessment, diagnosing, deciding, planning, communicating, managing, directing, and collaborating are viewed as macro-cognitive versus, for example, micro-cognitive processes such as noticing, managing attention, accessing information, or assessing options. In this conceptual scheme, the micro-cognitive processes support the macro-cognitive processes when deliberately designed to do so.
Decision-Centered Design progresses through five phases; preparation (development of domain understanding for the research group), knowledge elicitation (identification of essential work-related cognition), analysis (isolation of leverage points for supporting work-related cognition), design (development of a design concept), and evaluation (impact estimate of the proposed design). It relies on context specific, incident-based narratives to isolate leverage points for supporting the macrocognition involved in challenging situations . Decision-Centered Design identifies the key cognitive challenges and key elements of expertise involved in cognitive activity as a basis for generating design ideas that can support challenging cognitive work . Design solutions may incorporate one or more of technological innovations, work process enhancements, or cognitive skills training .
The creative design work undertaken within the framework of Decision-Centered Design is informed by views that align with Rasmussen’s problem-solving theory of cognitive modes . A cognitive mode is a style of cognitive processing used to undertake cognitive work. Although Rasmussen does not refer to macrocognition as such [30, 31], the types of cognitive processes he refers to align with those identified as macro-cognitive by those who promote Decision-Centered Design [25,26,27].
Rasmussen offers three modes of cognition; skill-based, rule-based and knowledge-based [30, 31]. The skill-based mode has no conscious processing between perception and action, the rule-based mode is guided by sets of procedural instructions that specify sequences of actions, and the knowledge-based mode is grounded in conscious and explicit reasoning. Identification of the modes used in any cognitive work guides the design of supports for that work, although a cognitive modes analysis needs to identify not only modes in use but also the information that supports the work and the type of action to be taken (Table 1).
Cognitive work such as diagnosis might employ any one of the cognitive modes on their own or two or three in combination. A clinician who recognizes sepsis at a glance is working in the skill-based mode. One who consults a checklist is working in the rule-based mode. One who notices the signs but must consult a text book to resolve what they mean is working in the knowledge-based mode. It is also of value to consider why different workers use different modes for the same task. For this sepsis illustration, an experienced practitioner may prefer the skill-based mode, an inexperienced practitioner may prefer the rule-based mode, and a student may prefer the knowledge-based mode. A cognitive support must conform to the cognitive mode or modes preferred by those who will use the system. Watson and Sanderson , in their development of sonic displays for anesthesiology, have observed that the cognitive modes supported by the monitoring technology do not align well with those preferred by anesthesiologists.
Several of the papers we cite in our background review note that clinicians often express concern about the impact information technology has on their workflow (the series of activities necessary to complete a task). As will become evident in our subsequent discussion, clinicians establish workflow patterns in part to support macro-cognitive processing. For example, a clinician who is concerned with maintaining their situation awareness within a clinical setting is likely to develop a workflow that will help them access the information critical for sensemaking in the desired sequence and at the right time.