Summary of the Case

Mrs. T., a 68-year-old woman with many cardiac risk factors and a history of myocardial infarction (MI), presented with atypical symptoms but a changed electrocardiogram (ECG). These ECG changes were not appreciated by Dr. M., the emergency department physician, and Mrs. T. was mistakenly sent home with what proved to be an acute MI.

Dr. M. was interviewed by a Quality Grand Rounds editor on 21 March 2002.

The Case

Twelve hours before presenting to the emergency department, Mrs. T. called the hospital's telephone triage nurse and reported "dull, midsternal pain relieved after a bowel movement." After probing for associated symptoms, the nurse reassured the patient and told her to call back if she experienced any further discomfort. Several hours later, when the pain recurred, Mrs. T. called again. When asked whether she had sublingual nitroglycerin on hand, Mrs. T. confirmed that she had a bottle of nitroglycerin pills but that the expiration date had passed. She was told to take the nitroglycerin if the pain recurred and was given an appointment for 2 days later (at which time she was instructed to exchange her expired bottle for a new one). She was advised to call 911 if the pain recurred and was associated with nausea, diaphoresis, or dyspnea. Because of continued pain, the patient came into the emergency department of a large urban hospital 4 hours later (at approximately 2:00 a.m.) with a chief symptom of chest pain.

The patient had a history of inferior-wall MI, hypertension, diabetes mellitus, hyperlipidemia, and peripheral vascular disease. She described the pain as "very different" from any pain that she had experienced in the past. It had a burning quality, was located across her epigastrium and chest, persisted for 4 to 6 minutes at a time, and had been intermittent for 24 hours. The pain came on at rest and was relieved by activity. She reported no associated dyspnea, diaphoresis, or nausea. Review of systems revealed 1 week of constipation.

Dr. M., a moonlighting internist, was awakened from sleep to evaluate the patient. Physical examination revealed a pulse of 85 beats/min, blood pressure of 140/70 mm Hg, and respiratory rate of 18 breaths/min. The lungs were clear to auscultation, and heart sounds were normal, with no rubs, murmurs, or gallops. The ECG (obtained during a painful episode) was interpreted as sinus rhythm, with normal axis and normal intervals ( Figure 1 , top). Dr. M. noted a Q wave in lead III but specifically noted the absence of ST-segment and T-wave changes consistent with ischemia. The patient was discharged from the emergency department 1 hour later with a diagnosis of chest and abdominal pain secondary to constipation. She was prescribed a regimen to relieve constipation and was told to schedule a follow-up appointment with her primary physician.

View larger version (102K): [in this window] [in a new window]   Figure 1. Electrocardiograms (ECGs) obtained at presentation and a previous comparison tracing.Top. The ECG shows tracing obtained at initial presentation (and obtained during an episode of chest pain) at half-standard scale. Bottom. The ECG shows tracing at standard scale from several years before the current presentation. It was available for comparison at the time of presentation, although the difference in ECG standardization was not noted at that time despite the presence of standardization marks. Subsequent recognition of this difference between the 2 tracings presumably contributed to the cardiologist's interpretation of the new ECG as abnormal, with new J-point depression anterolaterally (V2 to V5).

Diagnosing Chest Pain in the Emergency Department

Few diagnostic decisions have been more heavily researched than the approach to the patient with acute chest pain. In the context of the patient safety movement, it is useful to consider this case not only for what it teaches us about triaging patients with acute chest pain but also for what it may reveal about improving the individual physician's diagnostic performance through the use of algorithms or protocols.

Chest pain accounts for about 5.6 million emergency department visits annually, second only to abdominal pain as the most common reason for an emergency department visit. Approximately 1% to 4% of patients who present to an emergency department with what is actually an acute MI are mistakenly discharged (1-8), and the percentage of missed diagnoses increases when the denominator includes not only acute MI but also unstable angina. Patients discharged from the emergency department with MI have a generally worse prognosis than do appropriately hospitalized patients with MI (1-4), partly because of their risk for sudden death but also because of the delay in implementing treatments that are known to be effective for MI or the acute coronary syndrome (unstable angina or non–ST-elevation MI). Patients with atypical symptoms, and especially patients without chest pain (2, 3), are most likely to be mistakenly discharged.

The clinical question is which patients with acute chest pain have a presentation benign enough to make discharge from the emergency department safe and appropriate. Cost-effectiveness analyses suggest that a coronary care unit is the appropriate triage option for patients whose probability of acute MI is about 20% or higher (9, 10). For patients whose risks for MI or acute coronary ischemia are lower (5-8, 11), admission to telemetry units is often recommended, including a short stay on a chest pain (or coronary) evaluation or observation unit. In analyzing Mrs. T.'s presentation, it is essential to determine whether any combination of initial symptoms, signs, laboratory studies, or ECG findings has enough discriminatory power to reduce the likelihood of misdiagnosing an acute coronary syndrome to a level that would render discharge from the emergency department safe and appropriate.

In the acute setting, the ECG is not only the most important piece of information (12), it is nearly as important as all other information combined. About 80% of patients with acute MI have an initial ECG that shows evidence of infarction or ischemia not known to be old (Figure 2), and any patient who has such abnormalities has too high a risk to be safely discharged, regardless of the clinical history or physical examination (13, 14). The sensitivity is lower if the goal is to identify ischemia in addition to infarction, but comparisons with previous ECGs can improve the accuracy and usefulness of interpreting the ECG (15). Although a normal ECG at presentation predicts a relatively lower risk for complications (16-18), it cannot absolutely exclude myocardial ischemia or even MI. For example, among patients mistakenly discharged from the emergency department, up to 50% have normal or nondiagnostic ECG findings (2, 19). Thus, even if Mrs. T.'s ECG had been normal or unchanged from her previous ECG, this would not have had enough negative predictive value to exclude an acute MI or the acute coronary syndrome.

View larger version (20K): [in this window] [in a new window]   Figure 2. Receiver-operating characteristic curve of the initial electrocardiographic interpretation. The cumulative sensitivity (number of patients with myocardial infarction [MI] divided by all patients with MI) of electrocardiographic interpretations is on the y-axis and is plotted against the cumulative false-positive rate (number of patients without MI divided by all patients without MI) on the x-axis. New is defined as not known to be old. Reproduced from Rouan et al. (13).

The description of the presenting symptom is also important. Patients with chest pain are more likely to have MI or the acute coronary syndrome (7, 8, 11, 14), but up to 25% of patients with these diagnoses may present with symptoms such as shortness of breath, dizziness, or weakness, so cardiac ischemia must also be considered in patients with these symptoms. Demographic factors and traditional cardiovascular risk factors (with the very notable exception of a history of MI or coronary disease [5, 6, 20, 21]) are of little importance in predicting the cause of acute chest pain (21-24). Aspects of the medical history that appreciably lower the patient's likelihood of ischemia (likelihood ratios of approximately 0.2) include reproducibility of pain with palpation or positional changes, pleuritic pain, stabbing pain, or pain radiating to the lower extremities (5, 6, 20, 21, 24). However, even these negative predictors cannot reliably exclude MI (20, 25). Mrs. T.'s description of painful episodes lasting only 4 to 6 minutes may also seem "atypical," but the duration of symptoms is not a useful predictor (5, 7-9) unless the pain has persisted for 48 hours or more without ECG changes (5, 6). Patients who describe their pain as similar to previous episodes of cardiac ischemia are in a high-risk category (5, 18), but any chest pain carries a higher risk than no pain (7, 8, 11). Although the precise reproduction of chest pain by local palpation decreases risk (5, 18), normal results on physical examination do not lower the risk (5, 18, 20, 24).

How can these data have been used in caring for Mrs. T.? She had pain that was different from her previous MI and was thought to have an unchanged ECG. If it is assumed that all of these data are accurate, she would have had less than a 7% risk for MI and a low risk for complications that would require intensive care (18). However, because of her history of coronary disease and the absence of a clear-cut benign diagnosis, because constipation is not an established cause of chest pain, and because her pain had not resolved, she is the type of patient for whom admission to a chest pain evaluation unit is appropriate (20, 21, 26-33) (Table). It is very important for individual hospitals to adopt clear guidelines for triaging such patients, because these patients may be evaluated by many different physicians with varying experience, knowledge, personality traits, and levels of fatigue (8, 34, 35).

Dr. M.: I think one of the factors that affected my decision making when I first evaluated the patient was the time of night (2:00 a.m.) and the fact that I had just awakened. I saw her less than a minute after being awakened. What I probably should have done was had her stay in the emergency department, even if I thought she was low risk (which I obviously at that time did), and let more time pass so that my sleep inertia could wear off.

Many studies have documented the effect of fatigue on the cognitive performance of physicians (36) and other workers (37). Although fatigue is clearly a safety issue (38), strategies to reduce fatigue must be planned carefully to avoid adverse collateral effects. For example, a before–after study of the effects of regulations on residents' work hours in New York after the Libby Zion case revealed no change in hospital mortality after implementation of the new regulations but found delays in test ordering and an increase in complications (39), presumably because of the increase in handoffs. Improving the sign-out and handoff of patients can reduce the risk that discontinuity will worsen the quality of care (40, 41).

The Case, Continued

Mrs. T. was discharged from the emergency department. Three hours later (5:00 a.m.), she called the telephone triage service for her doctor's practice and informed the nurse that she had been seen in the emergency department for chest pain during the night, at which time her ECG was "normal" and the diagnosis was constipation. She added that she had had 2 bowel movements after taking milk of magnesia, but her symptoms had persisted. The nurse advised her to obtain a new bottle of sublingual nitroglycerin and to return to the emergency department should the pain continue or should new symptoms develop.

The Telephone: False Reassurance?

Recently, the availability of beneficial therapies for acute coronary syndromes has led to several initiatives to reduce delays between the onset of symptoms and the provision of definitive care. Given the importance of the ECG for both diagnosis and prognosis, triage systems should recommend immediate in-person evaluation of virtually all patients with a presenting symptom of chest pain, preferably in an emergency department (20, 42-45). Attempts to contact health care providers can delay the care of patients with the acute coronary syndrome by delaying emergency medical services and the arrival of such patients in the emergency department (46, 47). In a recent survey of primary care after-hours telephone-answering services, physician reviewers reported that patients were frequently asked to decide for themselves whether problems were emergent enough to contact the on-call physician and that 50% of calls not forwarded to the physician were emergencies that warranted immediate contact (48). A patient who does not improve after initial treatment may have received the wrong diagnosis and treatment and therefore deserves re-evaluation (42-45). In this case, well-meaning triage nurses seem to have relied on the patient's willingness to be reassured rather than insisting that the patient come to the emergency department for immediate evaluation (in the call 12 hours before the emergency department visit) and then relied on the "clean bill of health" from the emergency department physician in reassuring the patient (in the call after the emergency department visit).

The Case, Continued

At 7:00 a.m., approximately 2 hours after the second call to the telephone service, the patient returned to the emergency department where she was again assessed by Dr. M., who was now approaching the end of his shift. Before going off duty, the physician ordered a "GI [gastrointestinal] cocktail," screening hematology and chemistry tests, a troponin I level measurement, chest radiography, and electrocardiography. At 10:00 a.m., a new physician came on shift and noted that the patient had no symptoms. One hour later, however, the results of the troponin I test came back at 2.0 µg/L (normal level, 0.3 µg/L), thus confirming the diagnosis of MI.

The patient immediately received a chewable aspirin and intravenous metoprolol and was placed on a cardiac monitor. The note from the new physician reaffirmed Mrs. T.'s symptoms and reported normal results on physical examination, but now Mrs. T.'s new ECG was interpreted as showing pseudonormalization of the T wave in V ₆ compared with a previous ECG in 1998 ( Figure 1 , bottom). The patient was admitted to the cardiology service with the diagnosis of acute MI, although the physician's note specifically called the presentation "very atypical." An attending cardiologist later reviewed the ECG from the 2:00 a.m. visit to the emergency department ( Figure 1 , top) and interpreted it as abnormal, with new J-point depression anterolaterally (V ₂ to V ₅ ), compared with the previous ECG. In fact, the ECG showed about 0.5-mm ST-segment depression on half-standard voltage, roughly equivalent to 1 mm on a regular standard ECG—a finding consistent with ischemia compared with the baseline tracing.

The Elusiveness of Diagnostic Perfectibility

The modern approach to patient safety emphasizes "systems thinking" rather than individual cognitive mistakes or technical "slips." The goal is to create processes and solutions to prevent human errors, which are commonly made by competent individuals (49, 50). Suggested approaches include diagnostic protocols and pathways, decision aids, novel approaches to staffing, and other systems changes.

Several factors have been associated with inappropriate discharge of patients with chest pain from emergency departments: younger age, female sex, non-white ethnicity, atypical symptoms, no previous MI, and normal or near-normal ECGs (1-4, 51-57). Since about 25% of patients who are mistakenly discharged are sent home because of an error in the interpretation of their ECG (1-3), improving the analysis and interpretation of ECGs could improve decision making (8, 58-62). In this case, more accurate interpretation of the initial ECG might have prevented the patient's discharge from the emergency department.

Can further training or the use of decision algorithms help solve the problem of missed ischemia or MI? Standardizing ECG interpretations by traditional computer analysis (8, 63-65) or artificial neural network technology (66-68), and incorporating these interpretations into clinical decision making, theoretically could reduce diagnostic error. Several strategies have been used to attack this problem. Pozen and Selker and their respective colleagues (7, 8, 11) have developed models to predict coronary ischemia (encompassing acute MI and the acute coronary syndrome) because of its broad clinical relevance, whereas Goldman and colleagues (5, 6) tried to predict the probability of MI because it is more precisely defined and carries the highest risk. Eventually, Goldman and colleagues (18) switched to predicting the need for intensive care or other levels of care because it seemed to be the critical triage decision facing physicians. Early data demonstrated that physicians with higher levels of training had a higher sensitivity for detecting MI, but at the expense of decreased specificity—consistent with a shift along the same receiver-operating characteristic curve (34). Although several decision aids have good predictive power (5, 8, 18, 21), implementing them in clinical practice has generally demonstrated little or no overall improvement in triage decisions (69-78), except by trainees (8). Low-intensity interventions, such as chart reminders and guidelines, have been difficult to implement (76, 78). However, a predictive model with an automatic ECG interpretation can increase the use of fibrinolytic therapy for acute ST-elevation MI, especially in historically undertreated patients (79). Alternatively, Cook County Hospital in Chicago, Illinois, developed a consensus to adapt and implement Goldman and colleagues' (18) intensive care triage algorithm (75). This resulted in a highly statistically and clinically significant 15% absolute reduction and 39% relative reduction in the triage of uncomplicated patients to unnecessarily intensive settings without a corresponding reduction in the admission of patients with complications to appropriately intensive settings. These 2 approaches—incorporating validated predictive instruments into the ECG reading, or using validated predictive models to develop and implement local consensus guidelines—deserve more widespread testing to determine their broader generalizability.

However, any algorithm that incorporates only clinical elements and the ECG at presentation is likely to be suboptimal because of the substantial proportion of patients who present with atypical symptoms or no or minimal ECG changes. Widely available biomarkers (such as creatine kinase–MB assays and troponin T and troponin I levels) have low sensitivities for the diagnosis of MI when measured at the initial presentation to the emergency department, particularly within 6 hours of the onset of symptoms (12, 80). Although normal levels of these biomarkers cannot exclude unstable angina—by definition, a condition without myocyte necrosis—the finding of serially negative biomarker levels at presentation and 6 hours later identifies patients at low risk for further complications, particularly when combined with a normal or nondiagnostic ECG (12, 81, 82). Whether Mrs. T.'s initial troponin level would have been positive when she presented 12 hours after the onset of her pain is unknown. The key issue for diagnosis is not that an initially positive biomarker level is unhelpful—in fact, it mandates admission—but rather that the decision to measure and assess a biomarker level at presentation implies that the patient's risk is high enough to require a repeated assessment 6 hours later if the first level is normal.

Studies of immediate exercise stress testing, echocardiography, and routine predischarge coronary angiography have demonstrated good negative predictive values in low- to moderate-risk patients who have symptoms consistent with acute cardiac ischemia (51, 83-88). In a large randomized trial, prompt resting sestamibi scans statistically significantly reduced unnecessary admissions without changing appropriate admissions when used in patients with suspected ischemia and normal or near-normal ECG findings (89). However, most of these strategies have excluded patients with a previous MI and none have been widely adopted.

The sobering bottom line is that 2 decades of research has taught us that without compelling evidence for a noncardiac cause, there is no absolutely fail-safe way to exclude myocardial ischemia or infarction at the time of a patient's initial presentation. A short period of monitoring and measuring serial biomarker levels in a chest pain evaluation unit is an attractive approach for patients with an uncertain diagnosis. Although there may be substantial variations in the rates at which MI or the acute coronary syndrome are missed at individual hospitals (2), the "missed MI rate" is generally inversely proportional to the "rule-in" rate (90). Observing more patients in chest pain units can reduce the rate of missed MIs to less than 1% (90) and still allow expedited diagnosis and evaluation of a large number of low-risk patients. Rapid protocols and chest pain units are cost-effective alternatives to regular hospitalization for low-risk patients (26, 31, 32, 91, 92). Many experts recommend routine exercise or other provocative testing after 6 hours of observation, but data to support this practice are limited. Chest pain units can incorporate educational and community outreach programs focused on cardiovascular disease (27, 28, 32) and can improve patients' satisfaction (29). Chest pain units are also an ideal location to study decision algorithms and new strategies for diagnosis and therapy (93). These specialized units can standardize the clinical assessment and the interpretation of the ECG by training and employing a limited number of personnel. Given the profound adverse clinical outcomes and high malpractice awards associated with missed MI, chest pain units increase the likelihood that patients are truly safe for discharge.

When Dr. Goldman, the lead author of this paper, first began researching patients with acute chest pain more than 20 years ago, he believed that he and others could identify a set of clinical and ECG findings that would more accurately identify actual MIs than the physicians who actually saw these same patients. However, the ensuing years of research (18, 94) have revealed that there is no single way to discriminate perfectly between those who should be admitted to a coronary care unit and those who should be sent home. Instead, there is a continuum of risk—a finding that suggests that there also should be a range of triage options to match not only the diversity of risks but also the differences in the levels of care and monitoring among individual hospitals (32). From this appreciation came the expanded use of step-down units, beginning with their original focus on moderate-risk patients (9) and then, with the development of chest pain (or coronary) evaluation (or observation) units, a refocus on patients whose risks were even lower, yet not low enough to ensure the safety of immediate discharge (26-33, 90-93). What we have done, in essence, is to take the dollars saved per patient from these shorter observation protocols (as compared with the resource intensive, multiday "rule-out MI admission" of the past generation) and used them to admit more low-risk patients, thereby decreasing the missed MI error rate (90).

The Response to the Case

Dr. M.: I told the patient, when she came back about 3 hours later, that she was in fact having an MI, that she probably had been having an MI 3 hours earlier, and that I had misread her ECG. ... I apologized to her. She did not think that there was much harm from the 3-hour-or-so delay and was very forgiving. ... My junior colleagues were uniformly appreciative of my sharing the error, and they told me it helped create an atmosphere in which they felt more comfortable discussing their own errors. Our chairman was very supportive. He said that he was proud that I had been willing to discuss the error openly and hoped that this kind of disclosure would continue throughout our entire department. I think when we think about systems issues around errors, it is easy enough to say that we shouldn't point fingers at individuals and we should look at the system overall. While I felt that this error was in some ways personal, I don't think of myself as an incompetent physician. Unfortunately, ... beyond identifying, in retrospect, that the abnormality in the first ECG was probably underappreciated, I don't think that we, as a group, came up with any solutions. Again, I think the main kinds of solutions could be better training for me or other people in my position, possibly having other providers around who could have evaluated the ECG. But it was hard to come up with a simple solution of how we could prevent this kind of error in the future.

Honesty as a Trigger for Change

"If it ain't broke, don't fix it" is a popular aphorism in response to decisions about allocating scarce resources. Identifying and eradicating suboptimal care, assessing problems honestly, and committing to improving both personal performance and systems of care are cornerstones of a physician's responsibility to individual patients and to society. Recognizing an error and sharing the error with the patient and other physicians affords us and others the opportunity to recognize that errors do indeed occur, and that we can strive to become better physicians by learning from them. Mistakes must be opportunities for progress, not punishment.

Appendix: Questions and Answers from the Conference

An audience member: What do you think of the use of telephone triage in this case?

Dr. Goldman: Telephone triage can be useful when it is accompanied by well-constructed protocols and guidelines. My guess is that the nurses performing the telephone triage in this case lacked a chest pain protocol, which should have led them to direct virtually all patients with chest pain to come to the emergency department, since there is no combination of clinical variables that trumps a test that can be performed only in the emergency department, namely the ECG.

Dr. Robert M. Wachter, Quality Grand Rounds Editor (Moderator): The evolution of your thinking on this issue is fascinating—you have moved from a hope that we could perfect a diagnostic algorithm in the emergency department to a recognition that the best way to prevent errors is to admit most low-risk patients and then exclude the possibility of cardiac ischemia as efficiently as possible. Is any of this experience generalizable to the management of other disorders, or is chest pain unique?

Dr. Goldman: One lesson, which is probably a good one for all of us, is to remain humble and open to changes in our thinking. The chest pain experience is most generalizable to only a few disorders, such as pulmonary embolus, meningitis, or an acute abdomen, that also have such high stakes. For example, one could try to use the Pneumonia Severity Index to discharge low-risk patients to home (95, 96), because for this and many other diseases the adverse consequences of an inappropriate discharge are far less than with chest pain.