Error Reduction in the Emergency Department
Adoption of Error Reduction Techniques From Other Industries to
Emergency Department Procedures

     The practice of Emergency Medicine is a complex, variable and difficult job. Efforts have been made to analyze it according to standard industrial models, and with varying success. It is at the most complex end of the spectrum of systems humans operate, and therefore most error-prone. And unfortunately, errors do not result in unacceptable widgets being produced, they result in bad outcomes, including death, for our patients. No one argues that error reduction is not a noble goal. Many argue that systems this complex defy analysis by these industrial methods. However, in our current recognition of the appalling toll that errors cause, nihilism and objections to proposals to reduce errors are correctly viewed as self-interested resistance, and non-productive push-back. Analysis of similarly complex and life-or-death-dealing systems, such as operating room anesthesia, commercial airline operations, and U.S. Naval aircraft carrier operations, have shown that the analysis can be very fruitful in identifying risk points and areas for improvement.
     Emergency Medicine in a standard hospital setting is a complex system, in some ways linear, but in many ways functioning with multiple parallel processes interacting at various points, with various workers responsible for the processes, and with multiple outcomes possible, depending on the inputs from these parallel processes. It is also a high-risk, high variability endeavor, with multiple points at which error can be introduced, and with death and serious injury always lurking as a possible result of error. Strictly linear processes are amenable to the familiar Quality Control (QC) methods of auto manufacturing and similar assembly line processes, even if they are complex. Non-linear, complex, tightly coupled processes are not so transparent, nor as amenable to QC procedures, partly because the processes cannot all stop while awaiting a result or outcome of one particular step, and partly because the processes necessarily proceed based upon incomplete or even erroneous information, incomplete knowledge of the individual facts at hand, and even incomplete knowledge within the field in general about any specific problem encountered. Further, the process is necessarily time-limited and time-pressured. One source of time limitation is the urgency of the disease and the need to intervene. Another is the substantial risk, if a possibly urgent symptom is not investigated rapidly and the potential life-threatening diagnosis rapidly ruled out or ruled in. A third source is the need to process large numbers of people, most of whom have readily identifiable and non-threatening problems, and yet to sort out of these the ones who seem trivial at first blush, but actually harbor life-or-health-threatening problems not yet identified. And finally, there is just the simple pressure of patients wanting to ‘get in and out’, and who value their convenience and time over anyone else’s life-threatening disease, and are not the least shy in expressing their displeasure at having to wait.
Against this backdrop, it is not surprising that Emergency Departments are among the highest risk departments in most hospitals, in terms of claims filed and complaints lodged with the hospital, and in terms of payouts on malpractice settlements. It is also not surprising that pressure to limit these losses is becoming intense. What is surprising is how resistant to change the enterprise can be. Changes proposed from outside the ED are generally met with resistance based on maintenance of tradition, territoriality, and internal perception of ‘unique’ knowledge of the system which makes the proposed change ‘impossible’, or at least ‘impractical’. Proposals for changes from within the system meet internal barriers of hierarchy, in which doctors remain responsible for “medical care”, nurses for “nursing care”, and operations personnel for supplies, structural features, and support staff. The overall operations may not even be under one person’s or even one department’s control, although the department leadership is generally held responsible for outcomes, even in the absence of operational control. It is not that people want their system to fail, but they see their portion of the system as capable and functional, and see failures primarily elsewhere in the system. Almost all of the systems involved have evolved over time. The processes within them have been fixed by convenience at the time they began, or by limitations of architecture, or by limitations or personal preferences of staff, or by general industry practices based on limitations of knowledge at the time. They are also driven by consumer (patient and family) demands. Change comes mostly from analysis and reaction after failure. Such analysis traditionally has sought to assign fault, and often results in changes of personnel. Only recently has an emphasis on systematic cause of failure come to the fore. Changes which do occur are focused on the failure, and rarely encompass any more than is necessary to address the particular circumstance of the failure. The reaction of the department to the failure typically results in policy or procedure changes, which are promulgated at the time of the analysis, disseminated to the staff, placed in a manual, and gradually forgotten. The entire process or processes are rarely examined in full detail, and are generally not subject to radical re-alignment, barring some disastrous error, or other severe stress. Some changes do persist, and contribute to the evolution of the system, but after a few years and a few cycles of personnel turnover, there is rarely even institutional memory of why processes are done the way they are done.
     An analysis of how to do things better needs to proceed from an understanding of how we do things now. This alone can be a daunting undertaking. A typical ED receives patients in many ways. A patient may arrive by ambulance, with some assessment already accomplished, and some advance notice of their arrival and possible urgency. (or no notice at all) A patient may arrive at the triage window, and state a complaint either urgent or non-urgent. The complaint may have no bearing on why the patient is actually there, since all patients know that a complaint of “chest pain” will reduce waiting time, even if the real concern is something completely unrelated. A patient may only become a patient when they slip in the hall visiting another patient, and are brought through the back door by nurses on the floor. Patients stagger through the ambulance entrance after being stabbed or shot, and patients are occasionally born unexpectedly to other patients who only had abdominal pain as a complaint. Patients also become ED patients when the admissions staff has no bed available for an urgent direct admission, and they are redirected to the ED and “held” pending bed availability. This list is not all-inclusive.
     The next step in the process is supposed to be a sorting process, called “triage”, from military parlance. This is an example of practice from another area which was incorporated into general practice of Emergency Medicine, although its evolution subsequently has rendered it unrecognizable to a military practitioner of triage. The concept comes from military mass-casualty situations, in which battle casualties must be rapidly assessed and categorized into groups, in order to use resources efficiently to save the most salvageable, and to avoid wasting time and resources on unsalvageable cases. Triage involves almost no treatment, except momentary airway interventions or application of pressure to bleeding, which can salvage a soldier with a minimum of time. Triage also involves harsh choices, such as assigning a hopelessly wounded soldier to a category of comfort care only, so that minimal further time and resource is devoted to a vain effort. It also categorizes painful but non life-threatening injuries for deferred care, after the most urgent but salvageable cases are treated. This process, done correctly, preserves the maximum of life and limb under limited resources. It also translates poorly into the civilian world. First and foremost, the term applies to mass casualties. There is no need for sorting if only one patient presents. Nevertheless, in the ED the “process” is applied mechanically, more for administrative and record keeping purposes. Secondly, patients who are still alive, (and their significant others) generally expect that all possible interventions will be employed until the patient is actually dead, regardless of the apparent futility. Our ability to assess futility accurately is notoriously poor, so their expectation is actually not unreasonable. In the field, a gunshot wound through the chest with low blood pressure and with more than an hour of transport time to stabilizing care is for all practical purposes a dead patient, and with high reliability. A similar patient in an ED is already at the next level of care, if not definitive care. The use of the “unsalvageable” category does not work, except in the extremes of mass casualty rarely encountered on the civilian side. Third, individual patients among the “walking wounded” have almost no understanding of priorities of problems currently being managed, nor do they care. They are concerned with their own problem, and how fast it can be assessed and treated.
     Overlaid on this transplant from military practice is the need to gather demographics and medical information on the patient. True triage would bring the salvageable, urgent cases immediately to care without further delay, and to a certain extent, the civilian version functions in this way. However, in typical practice, “triage” has evolved to mean “record initial complaint, divert the obviously endangered to the back, and gather demographics and medical history on the remainder.” It has also devolved from a multiply-parallel process to a linear one, resulting in the complete undoing of its basic function. If the triage nurse is in the middle of an assessment, the assessment of the next to present will typically wait until the first patient is finished. Even if multiple nurses are available, the process becomes multiple linear parallel processes, limited at the front end. No matter how many personnel are available, such a structure will overload when the numbers of people presenting exceed the number of staff available to assess them individually. And the above process only encompasses the 80% or so of patients who arrive at triage. Other processes occur for ambulance cases and other portals of entry to the system.
     The stream of patients divides at this point in the process, between those who are deemed acute and are brought to the treatment area, and those assessed as stable enough to wait. Here there is variability in that if there are spaces in the treatment area, and no particular backlog in triage, non urgent cases may be brought back directly. Likewise, the assessment of what is urgent and what can wait is somewhat variable, depending on how full the treatment area is, and how urgent are the patients already being treated. A laceration of the hand with a blood-soaked bandage might be brought back immediately under normal circumstances, or wait an hour or more in the public waiting area under others. A patient who begins yelling and behaving in an inappropriately angry way might well be brought back out of triage sequence just to avoid a confrontation. In theory, the triage personnel assess the waiting patients periodically for deterioration of their condition, and might upgrade their category if appropriate, for more immediate treatment. In practice, their ability to do this is variable, with variability existing due to the numbers of patients and staff, and also variability due to the nurses’ individual skills and experience.
     Once in “the back”, a patient has vital signs re-assessed and recorded, and the remainder of a complete nursing assessment is gathered and recorded. This process might take five minutes and might take twenty-five, depending on complexity and also nurse variability. During that time, the physician might or might not even be aware that a new patient is in the back, and would generally have no idea of the nature of the problem unless a nurse requests that the doctor come immediately. This is generally done for serious problems, such as chest pain, but again, variability exists between nurses, based on skill and experience, and also occurs based on the other demands on the physician’s time at that particular moment. For example, if the physician is scrubbed and finishing a suturing procedure, the nurses might not interrupt for anything short of a “code”, even though in terms of acuity, a stable suturing procedure can be safely, though not conveniently interrupted. Variability also occurs based on physician openness to being interrupted, and on their individual tendencies to encourage or discourage verbal consultation by the nurses.
     Once the preliminary matters are accomplished, the physician portion of the encounter begins, with a history and physical. The physician usually carries a clipboard with the initial complaint and some aspects of the patient’s past medical history, medicines and medical allergies, as well as most recent vital signs. She does a fairly focused and problem-limited assessment generally, and then writes orders on the chart, or enters them in a computer in some cases. Most typically, however, written orders, either hand written or from checklists of common orders, are handed off to a secretary, who enters them into the computer, and then gives the nursing related orders to the nurse. There might or might not have been a conversation between the doctor and nurse about these orders, depending on the availability of each at the appropriate time. The doctor might have an opportunity to record the salient portions of the history and physical on the record at that point, but the need to deal with other urgencies may not allow that. Some doctors are themselves linear in their function, and insist on completing a task before taking on another. Some take on multiple tasks, and some have difficulties keeping all of them moving at once. Most of this variability is individual.
Orders awaiting nurse action are typically flagged, or placed in a rack awaiting nurse availability to accomplish. There may be a linear structure of one nurse generally assigned to one patient, (so-called ‘Primary Nursing’) or there may be teams, or sectors, or just any available nurse models in operation. Variability occurs here at multiple levels, with prioritization of orders dependent on a physician notifying the nurse personally, or consistent use of a flagging system, and dependent of course on the amount of work and the urgency of the other cases currently in the system. Such orders might include blood draws and establishment of intravenous access (IV), obtaining a 12-lead EKG, obtaining of specimens of urine or stool or sputum, initiation of oxygen, continuous electrocardiographic monitoring, and administration of various medications. Some of these tasks will be carried out by the nurse, some by technicians or nurse’s aides, and some by specialized technicians such as phlebotomists, respiratory therapists, and radiologic technologists. Some can be accomplished in any order, and some require prioritization, due to one task interfering with another, or being dependent on the prior accomplishment of another, or due to the particular needs of the patient. The physician may have specific desires as regards these priorities, and may communicate these desires, but the accomplishment depends frequently on the availability of staff, and the decisions ultimately made by the nurses and techs. Some of these orders and procedures are routine and predictable for various problems, and can be initiated on a protocol, or standing order. Some cannot, and must await individual physician or nursing assessment and individual order. Some orders, which could and should be done as standing orders, are not, due to nursing discomfort at initiating “orders”. And all such orders are dependent on the support for, or resistance to nursing initiation of orders, by the individual physician.  A single physician insisting on physician prerogative and questioning nurse initiated orders as they occur can scuttle the entire process.
     The next phase of care is the accumulation of data and the observation of the patient’s course after the initiation of initial therapy. Here the physician has relatively little contact with the patient, although she might check on progress intermittently, time permitting. The nurses are responsible for checking on progress, and reporting back to the physician. There is a large amount of variability here. Nurses may or may not see their role as gatherers and reporters of significant information. They may or may not check on patients, and update the physician. They may expect that the physician finds all the results and reacts to them, without any input from the nurse. When lab results become available, and are grossly abnormal (Panic Values), these results are usually brought to the physician’s attention, and reaction to individual results might generally take the form of an order for additional therapy, or more testing. These orders might be written, discussed verbally with the nurse, or both. Abnormal results not of that level of concern might or might not be brought to attention, as multiple results are released at once. They may be available on a computer, but not brought to notice until the physician specifically queries the system for the results. Variability in these results reaching attention occurs here quite commonly, and with a high potential for error resulting. Variability in nursing practice as regards periodic updates to the physician also occurs, with significant information going unreported, or alternatively being buried unrecognized in a mass of trivial information, and mechanically reported without adequate prioritization.
     Finally, after accumulation of data is complete, and therapeutic trials initiated and results assessed, the time for disposition has come. The physician must review all the data, confer with the nurse, perhaps reassess the patient, and then determine that the patient must be admitted, or transferred to a higher level of care, or can safely be discharged. The physician must then contact an admitting physician, or make a plan for outpatient care. She must then put in an order for admission, or write instructions for after care and follow up, along with prescriptions, work excuses, orders for medical devices, and orders for follow up appointments or testing. These are all reviewed with the patient and family, either by the physician and nurse, or at least by the nurse. Normal practice is to verify understanding by the patient or caregivers, and to document this understanding. At that point the visit is concluded, either by transfer to the floor, or by discharge. There remains only a substantial amount of paperwork to document all aspects of the encounter, and a reassessment of all other tasks then pending, along with the status of the next presenting patient. That there should be error in the above outlined processes should surprise no one. That it ever occurs as intended and without error is the more dubious proposition.
     A similarly complex system which provides useful parallels is the U.S.Navy’s carrier operations. In that example, highly complex operations requiring much precision, with high risk for fatal error, are carried out under difficult and variable conditions by a staff with rigid hierarchy and high turnover rate, but to a very high degree of reliability. Each of these features bears similarity to ED function, and each would be expected to add a degree of uncertainty and dysfunctionality, and yet the Navy carries out thousands of launches and landings per ship, with millions of individual component actions carried out in sequence, by tens of thousands of individuals, and all at vanishingly low error rates we could only envy in the medical world. Several methods of error reduction employed by the Navy have immediate application to the ED, but the Navy is by no means the complete example. Civilian aviation has parallels as well, with millions of operations yearly, and a very low fatal error rate, despite the constant possibility of error leading to mass casualties, rather than resulting in a single fatality, as in medicine. And manufacturing industries have lessons to teach us also, in examples of work teams, automation, and information sharing. A variety of possible strategies can be culled from all these sources, and applied to Emergency Medicine, with the expectation that we can reduce error and injury substantially. Given the record so far, we cannot afford to remain complacent.

Redundancy
Redundancy is a concept not unknown in medicine, and it is employed to advantage in error reduction already. It is not employed systematically however. There are two modes of redundancy, namely procedural redundancy, and physical redundancy. Both can be employed to good effect. Procedural redundancy is the use of two people to accomplish a task, utilizing two sets of eyes and experience bases to cross check a safety critical procedure. Physical redundancy is the acquisition of two pieces of equipment or items of supply, so that in the anticipated failure of one, a second is immediately available. Depending on criticality and likelihood of failure, more than single layer physical redundancy might be required.
     An apt military example of procedural redundancy has been termed “stressing the survivors.” Under normal circumstances, multiple personnel have overlapping areas of responsibility, but none is at maximum capacity under normal operating conditions. There is a sufficient surplus of personnel that part of the duty of each is to observe the ongoing functions around them for which they are partly responsible, and to provide safety crosschecks and input as abnormal events are observed. If there are losses, or if the system is saturated and maximally stressed, the surviving personnel will have to step up to maximum capacity function, and some of the redundancy will necessarily be lost, or assumed by assets outside the immediate unit. The assumption, however, is that there is an expected “inefficiency” built into the routine staffing level, which allows procedural redundancy, and to a certain extent, physical redundancy of staff.
     Procedural redundancy in health care is employed currently in certain medications. By long and oft-quoted tradition, doses of insulin are drawn up by a nurse, and a second nurse is required to check that the syringe contains the written dose of insulin, before it can be given to the patient. This is a perfectly rational and useful example of increased safety achieved by procedural redundancy. Its origin is lost in the mists of time, but the behavior is embedded in nursing training nationwide, and is reproduced broadly in both hospital policies and procedures, as well as in actual practice. At the same time, the same nurse is free to draw up other, more frighteningly lethal medications on a scribbled order or verbal order from the physician, and inject these medications intravenously without ever a thought of cross checking. Thrombolytics, various cardiac medicines, antibiotics, chemotherapy agents, and a host of others are not subject to the same institutionalized scrutiny. Individual nurses do cross-check meds on an ad hoc basis, but this is driven by their familiarity with the med, their comfort with the ordering physician, the time pressure, etc. Interestingly, doses of potent narcotics can be drawn up and administered without cross check, but any instance of a partial dose being given and the wastage of the remainder requires cross checking, a practice driven not by patient safety, but by control of potential narcotic diversion. The systematic use of this cross checking could reduce one of the commonest identified sources of significant error, medication errors. Barriers to its employment include the necessary demand on staff time, which of necessity reduces availability for other tasks. This demand will ultimately require more staff, and therefore more salary expenses. A cost-benefit analysis will likely show that this is nevertheless cost effective, because it could impact the single largest source of error in the ED. At the physician level, redundancy is harder to achieve. In single coverage, there is no one available to double check decisions, nor is the practice warmly viewed in the field. Doctors are expected to know what to do, and though they do consult one another, as a generalist consults a specialist, constant cross checking within the same level of care tends to be viewed as weakness, or lack of knowledge. An area of redundancy that is not exploited in medicine is the practice of cross checking by any level in the work team. A technician questioning a doctor’s order, pointing out to Nurse B that Nurse A was just in the room giving potentially the same medication might or might not be warmly received. The Navy uses this technique to great advantage, by allowing flight operations to be suspended by anyone on the flight deck, if a safety issue is noted. While the person might be criticized if wrong, he is never punished, and is rewarded when correct. Here hierarchy gives way to safety, and medicine could use the lesson. In fairness, the most functional work teams in medicine do use this paradigm, and the most functional team leaders are often the least rigid in terms of role, but the practice is not standardized nor even well recognized.
Physical redundancy is also well known in medicine, but underutilized. In critical areas, such as the OR, ER and ICU, certain obviously critical devices have designated backup devices already present. Most of these are kept in reserve by established procedure, and many of these procedures represent the result of retrospective analysis of actual failures in the past. Prospectively, not many hospitals have taken the step of analyzing each procedure in each area, asking the question of where is the designated backup if (X) fails. Furthermore, expecting staff to find a manual to identify the backup plan at the time of a failure is designing failure into the system. The location of each backup needs to be posted on each device house-wide, so that at the time of failure, a manual does not have to be located and consulted. In many cases, the result of any specific equipment failure may not immediately be safety critical, but will impact the organization elsewhere if a backup is taken from another department, or cannot be obtained at all for a prolonged period. This may result in an unanticipated safety issue elsewhere. Physical redundancy may also be as simple as insisting that an array of endotracheal tubes already be available and assembled for use prior to intubation, in case the primary tube is dropped, or is too large, or the balloon is violated on insertion. A systematic evaluation of each department’s procedures with prospective failure analysis is required to see where safety gains can be made. Much of the redundancy already exists in terms of equipment and supplies, but the benefit is not available due to faulty procedures. These barriers can be reduced.

Checklists
Aviation provides several additional techniques which can be employed to advantage in the ED. One well understood technique is the use of checklists. Each airplane type has multiple procedures specific to its operation, and its operation is a complex, sequential task, with high potential for failure if procedures are omitted. Each aircraft has associated with it a series of checklists applicable for various phases of operation, both routine and emergency. Starting a simple single engine piston plane requires twenty or more procedures, and pilots are completely accustomed to use of a printed list to avoid errors of omission. No matter that the pilot has 10,000 hours in the type, and 3,000 takeoff and landing cycles under her belt, she still uses the checklist. Emergencies have multiple complex procedures for solving problems while continuing to navigate and aviate, and pilots, ideally with the help of redundant copilots, run these checklists to investigate and isolate the failure, while determining where and when to land. In Emergency Medicine, by contrast, we rarely use printed checklists. Much of physician training involves memorizing lists of things to check and sequences of actions to carry out, and we are expected (and expect of ourselves) that we can produce these lists instantly on demand, without omission, and without reference to printed material. This is of course a crazy and unrealistic expectation, and the cause of innumerable errors of omission, but yet the attitude persists. Pilots live in a world of high regimentation along with high authority. The pilot is responsible for the safety of the flight, and may violate FAA rules and regs if dictated by safety needs, but pilots also comply with a degree of regulation in routine operation which physicians never face. They do so because of their culture, but they do so also because they are employed by airlines, or are officers in the armed forces, and because they are required to do so. Doctors are historically independent, although becoming less and less so. Nevertheless, their culture is anti-regulation, and resistant to admitting that errors of omission are the inevitable result of faulty human memory. An FAA of medicine is a far-fetched notion, but that level of regulation may come to medicine if medicine does not reduce error to acceptable levels independently. In ED processes, checklists could be very beneficial in preventing discharges before all ordered tests are returned and reviewed, or without ordered meds or treatments. The entire list of tasks at the end of the encounter can readily be assembled and reviewed each time, to ensure completeness, and to ensure that prescribed treatments do not conflict with known allergies, etc. Likewise, on intake, and depending on presenting complaint, a checklist containing standardized procedures can ensure that oxygen is started if appropriate, that monitoring is placed as soon as the patient is in the bed, that vitals are obtained and stable, that the call bell is available and the rails up as appropriate. This list is carried out ad hoc by the nurses currently, and errors of omission are a constant and unassailable presence in the ED. The proliferation of pathways or protocols which are disease specific represent another use of checklists. These have been developed piecemeal to deal with documented omissions in the care of the most common diagnoses. They do have the potential to increase the percent of patients who receive standard meds in certain diagnoses, such as aspirin and beta blockers in unstable angina and MI. They have unfortunately met with resistance from physicians, who deride them as cookbook medicine. Another example which can be helpful is standardized order sets, based on presenting complaint. Abdominal pain, for example, will almost always require a list of labs including urinalysis, blood count, basic chemistries, and a pregnancy test in females of appropriate age. The nurse in triage can initiate this sequence immediately, preventing delay in care by “rounding up the usual suspects” at the earliest possible time, and preventing omission of one or another step, preventing missed diagnoses, or at least further delay, while the omitted lab is obtained later. Multiple other complaints are amenable to this technique, and yet physicians continue to object that it eliminates their professional judgment. In truth, it will result in too many labs being done when trivial pathology with the same complaint could have been managed without lab at all. And to the extent that the physician is available to see patients as rapidly as they present, they are free to exert their judgment up front. But to the extent there is a waiting period before evaluation, the time, staff and ED bed congestion costs must be weighed against the occasional overuse of lab testing. A rigorous analysis has not been published, but the balance is likely in favor of standing orders.
Another aviation model is the concept of the “sterile cockpit”. FAA records are replete with examples of distracted flight crews carrying on unrelated conversations during critical phases of flight, particularly takeoffs and landings, and failing to note warning lights or sounds in time to avoid disaster. Similarly, “crew resource management” (CRM) procedures dictate that there be clear understanding between the pilots of who is flying the plane and who is jiggling the loose wire under the radio. Documented civil aviation crashes have occurred as both pilots became absorbed in a non-critical warning light, and neither flew the plane. In the ED, as in the cockpit, there is a culture of gallows humor which is part coping mechanism and part machismo. It is not unusual to have resuscitation situations, which are often futile exercises in any case, be remembered more for the witty repartee than for the outcome. Even without this phenomenon, there is constant, often loud chatter going on which may relevant to the tasks at hand, but which obscures the flow of information up and down the chain. There is variability in how effective team leaders are at recognizing the problem, and how they cope with it. Formal training in trauma codes includes assessment of and reduction of distractions, but the concept of sterile cockpit could be employed in many other areas. Avoidance of “task saturation”, with everyone in a code paying attention to one task to the detriment of their assigned task is likewise a learned behavior, reinforced by training, and seldom achieved individually without training.

Feedback Loops
Much of medicine is carried out in sequential input-output cycles. An order is placed, a lab is drawn, and a result reported. This result raises other possibilities to be investigated, or requires a therapeutic intervention. The doctor responds with the appropriate test order or med order. All of this depends on results reaching the ordering physician, and efficiency and safety dictate that they reach her as rapidly as possible, and as soon as available. This is the phenomenon of “feedback loops”. The sheer bulk of this information provides one problem, since a typical order set might contain 20-30 expected results, and require positive action to review each of them for relevance. Results are already prioritized by reporting abnormals in a highlighted format, and reporting defined “critical values” by direct verbal report to the ordering physician. However, the presence of a “normal” result in the setting of a suspected diagnosis incompatible with this result is not a scenario captured by such a protocol. For example, if a man with suspected kidney stone has a “normal” urine, when blood is expected in the urine, the routine normal report without urgency may not alert the physician to the fact that an abdominal aneurysm is actually the cause of the pain, and that the patient is about to expire without surgical intervention. Likewise, an abnormal result which is reported into a computer database, requiring physician action to access and review it may transmit the needed information, but information resident in a database is not the same as information in the possession of the doctor. This extra step can result in delay or omission, and does lead to error. In manufacturing and in aviation, critical results which determine the next action in a sequence are commonly presented in formats which “force” review and response. Alarms may sound which must be addressed before the next sequence can be initiated. Orders for an action might require that the precipitating data be included in the order. In medicine, this occurs with some regularity, but not with uniformity. Many lab results require urgent action, but because the result is separated in time from the order, the positive feedback portion of the loop is lost. A positive Pap smear reported two weeks later, by mail, in a stack of 50 negatives, has caused numerous preventable cases of cancer to go on unchecked. An abnormal potassium level, not highlighted in a list of twenty normal results reported routinely two hours later, could be the source of both severe injury and legal controversy. And as electronic data transmission has taken hold in the industry, many of the old reliable systems of paper-based notification systems have been dropped, but without adequate systems to replace them. In our institution, lab results formerly came as printed reports, with ranges of normal and abnormal included with each result. The physician got a piece of paper thrust into her hand, or placed on a chart for review, and presented by the secretaries. She reviewed, initialed, and responded to the results. Results not reviewed or not initialed were re-presented, and very few results were missed. That system did not capture results which were simply not reported, and these misses relied on the physician noticing that a desired result was still not available for review, and either tracking it down, or tasking someone with the chore. Currently, results are released into a database which the physician must affirmatively access to find results. If a result is not yet available, the physician must access the system again later to get it. If a result among many is reported earlier than others, it does not reach the physician until the physician seeks it. If there is a missed order, or a failure to enter the report, the physician must still note the absence of the desired result, and initiate an investigation to determine if the order was missed, or the blood not drawn, or just not run, or perhaps just not reported. The net effect is that the electronic medical record (EMR) is actually less functional in terms of feedback loops than was the paper system it replaced. Feedback loops can be designed into EMR products, but the problem must be understood by the programmers, and the effects on safety and functionality must be assessed by the caregivers. And the goal of forced feedback loops must be understood as a design priority in the first place. Currently the assumption is that the physician is responsible ultimately for results of orders, but it is clear that designing systems which require repetitive human action instead of utilizing the power of repetitive machine action is the same as intentionally designing failure and error into the system.
     Another feedback loop-related technique, which comes from aviation and military practice is “readback”. On a ship’s bridge, a command to steer a set course at a certain speed is repeated back to the commander by the helmsman. Ahead full, 2-7zero degrees AYE! A clearance from the Air Traffic Controller is read back verbatim to the issuing controller by the pilot. This serves the dual function of confirming that the order was even received/heard by the receiver, and that it was understood. These practices are so routine in aviation and military practice, they barely provoke any notice. In medicine, however, despite the historical love of hierarchy and chain of command, the practice has not taken hold. It has much to recommend it, but will require training and enculturation. With the new JCAHO standards comes the mandate that verbal phone orders must now be read back to the ordering physician. Physicians have grumbled and resisted, but the practice is unassailably correct, and will prevent many errors in exchange for a minimal expenditure of time. In the ED, particularly in code situations, but essentially always, verbal communications are complex and burdened with jargon. Many drugs have similar names, and numbers and letters are frequently mis-heard. Multiple patients with similar sounding names come and go in rooms with numbers. Errors are frequent, and many would be caught by adopting this practice. Here the cost will be training, and enlisting buy-in by the participants, since the cost in practice would be negligible.
Still another feedback loop which is not employed widely is the affirmative requirement for reporting results of interventions. If a respiratory treatment is ordered and delivered, the therapist is the only one then in possession of the information about what was the effect on the patient. She can complete the order and depart, or seek out the physician and report on effect, so that additional therapy can be ordered if needed, or the fact of improvement communicated. Likewise a nurse giving pain med can just carry out the order, or can reassess and report in twenty minutes. The cycle of input-output can be compressed, and more care delivered much more efficiently in this way. But this does require the training and expectation that an order presumes and requires a return report on effect. And the extra work for the nurses will require buy-in, so that they see benefits in terms of increased power and influence in patient care, as well as improved turnover of beds and operational efficiency. ED nurses don’t generally like long dwell times in their beds, when more patients are waiting in the waiting room. They do like to make a difference in their patient’s course and care.

Automation
The notion that automation can be the salvation for medicine in terms of the flow and speed of data is attractive, but the holy grail remains a distant vision. As noted above, automation can worsen rather than improve some situations. On the other hand, as these early problems are understood and designed out of future versions, the utility should improve. However, the potential in terms of error reduction is enormous. Just the elimination of handwritten notes and orders will be an enormous victory. Pulldown lists of available meds at available doses will speed drug ordering from manageable formularies. Drug orders will be entirely legible, and transmitted instantly to the pharmacy. Inventories can be maintained simultaneously, preventing shortfalls of supply. Med lists can be scanned and reproduced immediately, and updated by exception, rather than re-writing each individually. Allergies and incompatibilities can be picked up and flagged, and pre-screened alternatives suggested. On the other hand, the industry must be aware of new types of errors introduced by new technology. Excessively rigid menus may not allow legitimate but unusual uses of meds. Rapid reproduction of erroneous or outdated medication lists, and transmission to other sites of care can preserve errors on the original list, and place many copies of the error into the system. And of course, all databases are subject to human error during data input.

Human Factors
Human beings function best when they are well rested, focused on the task at hand, motivated, and properly trained. This happy confluence seldom occurs, and if it does, it does not persist for long. Error occurs even under optimal conditions, but the rate of error rises drastically as fatigue sets in, and error also rises as each of the other factors vary. A famous incident of resident fatigue during an on-call period of 36 hours leading to a young woman’s death has resulted in much discussion and even state regulation. The typical reactive response of the regulatory agencies has been followed by steady erosion of the changes, to the point that there is very little difference beyond official posture a decade later. In principle, everyone knows that doing complex tasks for 24 hours in a row results in reliably poor performance. In practice, the expense of hiring more staff to enable shorter shifts and prolong rest time has created resistance. The tradition in the ED is built around twelve hour shifts for both physician and RN staff in many locations. Multiple studies show that by the end of a twelve hour shift, the ability of a worker to remain focused on task is generally very poor, and error rates rise, as output speed declines. The “change of shift” phenomenon is observable every day, as throughput times rise, and patients wait longer, coincident with the outgoing shift’s unavoidable slowdown. A study of errors by time of day/relation to shift change would be interesting. A study of errors in a stable system before and after the adoption of a change from 12 to 8 hour shifts might also be very enlightening. What is clear, however, is that systems currently in wide use take almost no account of error probability, and are driven instead by the desire of the staff for time off, and by HR concerns and staffing costs. Other industries, more highly regulated than medicine, such as over-the-road trucking and commercial aviation, require strict adherence to limited work schedules, and mandatory rest periods before a new shift can begin. These measures are costly, but survive in these industries because error is costly and not tolerated to the extent it is in medicine.

Conclusion
The call for reduction of errors in health care has arisen as a relatively sudden outcry, prompted by the Institute of Medicine’s report "To Err is Human." Responses range from defensiveness, to an open dialog within the medical community, to politically driven regulatory responses. Medicine has much housecleaning to do, and nothing to gain from a posture of defensiveness. On the other hand, the health care industry does not operate in isolation, and as resistant as it might be to suggestions from outsiders, insiders can well afford to look to the outside world of other industries for examples of systematic error prevention which can be imported and adapted to medical use. The above examples amount to a catalog of the obvious, examples of “low hanging fruit.” Some are almost cost-free. Some will require commitment of substantial money and time. These demands will play out in our world of diminishing resources and rising costs, particularly for health care. But the IOM makes it clear that error is expensive, both in money and in lives lost or forever altered for the worse. It is also clear that error is inefficient. Some of the expenses required for systematic correction will be offset by reductions in error, and some will not. Some will just have to be borne as a cost of a system which no longer accepts routine error as normal.



References:

To Err is Human—Building a Safer Health System, Kohn, Corrigan and Donaldson, eds., Committee on Quality of Health Care in America, Institute of Medicine, National academy Press, 1999.
Dement WC. The perils of drowsy driving. N Engl J Med 1997; 337: 783-784.
Jones MA. Medical negligence. London: Sweet and Maxwell, 1991.
Leape, Brennan, Laird, Lawthers, Localio, Barnes, Hebert, Newhouse, Weiler, and Hiatt. The nature of adverse events in hospitalized patients. Results of the Harvard Medical Practice Study II, NEJM, Volume 324:377-384 February 7, 1991 Number 6.
Lowenstein, "Where Have All the Giants Gone?: Reconciling Medical Education and the Traditions of Patient Care with Limitations on Resident Work Hours"
Perspectives in Biology and Medicine - Volume 46, Number 2, Spring 2003, pp. 273-282
The Johns Hopkins University Press.
McCall TB. The impact of long working hours on resident physicians. N Eng J Med 1988; 318: 775-778
Nebeker et al., High Rates of Adverse Drug Events in a Highly Computerized Hospital, Arch Intern Med 2005;165:1111-1116.
Pierce, Promoting Patient Safety by Preventing Medical Error, JAMA 1999;281:1174-1174.
Powers MJ, Harris NH, Lockhart-Mirams A (editors). Medical negligence. 2nd ed. London: Butterworths, 1994.
Rochlin, La Porte, Roberts, The Self-Designing High-Reliability Organization: Aircraft Carrier Flight Operations at Sea., Naval War College Review in 1987, reprinted in 1998.
Rosen et al., Physicians' Views of Interventions to Reduce Medical Errors: Does Evidence of Effectiveness Matter? Acad Med 2005;80:189-192.
Schenkel, Promoting Patient Safety and Preventing Medical Error in Emergency Departments., Acad Emerg Med 2000;7:1204-1222.
Sexton, Thomas, Helmreich, Error, stress, and teamwork in medicine and aviation: cross sectional surveys, BMJ 2000;320:745-749 ( 18 March )
Shojania et al. Safe but Sound: Patient Safety Meets Evidence-Based MedicineJAMA.2002; 288: 508-513.