We evaluate a noncoercive, low-intensity intervention intended to improve the efficiency of care for patients with acute chest pain. This intervention provided information and recommendations to clinicians in the emergency department and hospital without person-to-person contact. Our findings may have implications for the application of practice guidelines.

Methods

Top Methods Results Discussion Author & Article Info References

Patients

Those patients aged 30 years or older who presented to the emergency department of Brigham and Women's Hospital between July 1990 and February 1992 with chest pain unexplained by local trauma or chest radiograph were eligible for the study. Data were collected for patients who presented during the intervention and control periods of this time-series trial. The six 14-week cycles each included 5 intervention weeks followed by 5 control weeks, separated by 2-week "washout" periods. These cycles did not correlate with rotations in physician schedules.

In the emergency department, clinical data were entered on a form by a physician at the time of clinical evaluation or by a research nurse as described in a published protocol [1]. The person completing the data form had no knowledge of, and thus could not be influenced by, the patient's subsequent course. Hospitalized patients were assigned at the discretion of the evaluating physicians to coronary care or intermediate care units, including a coronary observation unit adjacent to the emergency department [12]. Data about cardiac enzyme measurements, complications, and procedures after admission were recorded daily by study personnel.

Outcomes were classified by reviewers blinded to clinical data from admission. Acute myocardial infarction was diagnosed if characteristic evolution of serum enzyme levels was seen as follows: 1) creatine kinase-MB isoenzyme detected in more than trace amounts by qualitative electrophoretic assay or in amounts of 5% or more of an elevated total creatine kinase level with a typical increase and decrease by quantitative assays; 2) if creatine kinase-MB isoenzyme levels were not assayed, a typical increase and decrease of total creatine kinase level with a peak value at least twice the upper limit of normal; or 3) a lactate dehydrogenase isoenzyme one level greater than the lactate dehydrogenase isoenzyme 2 level in the absence of hemolysis or renal infarction. In patients with acute ST-segment elevation who received reperfusion therapy with intravenous medications or primary angioplasty, myocardial infarction was diagnosed if catheterization showed 100% occlusion of the infarction-related artery or if the creatine kinase-MB isoenzyme level was at least 2.5% of an elevated total creatine kinase level. Patients who had sudden cardiac arrests were classified as having acute myocardial infarction if they died before enzymatic confirmation of myocardial necrosis could be obtained and if there were no other explanations for the arrest.

Resource utilization data for all hospitalized patients were retrieved from the hospital's fiscal database. These data included length of stay and costs in 1992 dollars as estimated from charges using cost-center-specific ratios of charges to costs.

Intervention Design

The intervention provided risk estimates for patients at the time of emergency department evaluation and for the first three mornings after admission to the hospital. In the emergency department, physicians had access to a patient's probability of acute myocardial infarction as predicted by a slightly modified version of a validated multivariate algorithm [1] and to the patient's probability of major life-threatening complications as predicted by another algorithm [13]. During intervention periods, these algorithms were posted on walls and stapled to the backs of the data forms that were completed by physicians as progress notes. Physicians therefore had the opportunity to review the algorithms before completing the data form. Aside from education about the purpose of the study and the content of the algorithms, physicians did not receive direct contact encouraging them to use these algorithms.

For patients who were hospitalized, study personnel provided predictions of the risk for "complications requiring coronary care unit care" on colored stickers that were applied to the front of each patient's chart before 7:00 each morning. This estimate reflected the probability of cardiac arrest; ventricular fibrillation; cardiogenic shock; complete heart block; atrioventricular dissociation; Mobitz type II atrioventricular block; emergent intubation; emergency cardioversion; temporary pacemaker placement; intra-aortic balloon pump placement; pulmonary artery catheterization; and left-heart catheterization within 3 days of admission followed by coronary bypass or valve surgery during hospitalization. These predictions were based on clinical data from the patient's initial evaluation and subsequent course as analyzed using data from the Multicenter Chest Pain Study [13]. This end point was chosen because most patients without these complications and procedures can be managed with the nursing available at lower levels of care.

The stickers also gave recommendations about the appropriate level of care for patients with a given probability of complications. After the first intervention cycle, triage recommendations were added to the algorithms used in the emergency department in response to requests from physicians for guidance in interpreting quantitative risk predictions. For example, at 12 hours after admission, patients were considered to be at low [<2%] risk for complications requiring care in the coronary care unit if they 1) had less than a 7% risk for acute myocardial infarction according to data from the emergency department evaluation, 2) did not have abnormal cardiac enzyme levels, 3) had not had recurrent ischemic chest pain, and 4) did not have other cardiac or noncardiac conditions requiring coronary care unit care. Noncardiac complications requiring intensive or coronary care included respiratory failure leading to intubation; sustained hypotension (systolic blood pressure < 90 mm Hg); active gastrointestinal bleeding; and cerebrovascular accident. These risk prediction strategies and triage recommendations were reviewed by the director of the coronary care unit to ensure their acceptability to other physicians, and by a hospital attorney. The study design was approved by the Institutional Review Board of Brigham and Women's Hospital. Throughout the trial, the intervention avoided human contact as a method for transmitting patient-specific information.

Statistical Analysis

The primary end points of the trial were rates of admission to the hospital and coronary care unit, estimated inpatient costs, and length of stay. Univariate comparisons of continuous variables were done using the Wilcoxon rank-sum test. Univariate analyses of categorical variables were done using a chi-square test. A P value of 0.05 was used to define statistical significance. Because individual triage decisions were nested among individual physicians, we adjusted for possible intraphysician correlation using a Generalized Estimating Equation procedure [14] and logistic and linear regression models that also adjusted for the Charlson comorbidity score. Because the findings from the adjusted analyses were nearly identical to those from the unadjusted analyses, we present only the latter.

Results

Top Methods Results Discussion Author & Article Info References

The time-series design yielded intervention and control groups similar in size and clinical characteristics (Table 1). The prevalence of previous myocardial infarction was 23% in both groups, and the overall prevalence of electrocardiographic findings consistent with ischemia or ischemia not known to be old was 21%. The percentage of patients found to have acute myocardial infarction was nearly identical in the two groups. The only statistically significant difference was that the control group had a slightly higher prevalence of comorbid medical conditions as measured using the Charlson comorbidity index.

Comparison of Resource Use in the Total Sample

No differences in triage decisions or resource use were found between the intervention and control groups (Table 2). Rates of admission to the hospital and to the coronary care unit were nearly identical, as were total lengths of stay and estimated costs for patients hospitalized. There was a weak, nonsignificant trend toward shorter lengths of stay in the coronary care unit for the 131 patients in the intervention group compared with the 126 controls admitted to the coronary care unit (P = 0.32). This sample size had an 80% power for detecting a decrease in hospital admission rates to 45%, coronary care unit admission rates to 6%, and average length of stay to 3.9 days.

Comparison of Resource Use in "Low-Risk" Patients

Because a previous hypothesis was that risk information would increase the efficiency of care for patients at low risk for complications, we did secondary analyses among the 1160 patients whose risk for acute myocardial infarction was 7% or less according to their initial clinical data Table 3 [1]. Only 1.6% of these patients were admitted to the coronary care unit. No differences were detected in management for these patients between intervention and control periods.

Discussion

Top Methods Results Discussion Author & Article Info References

We describe the failure of a low-intensity, noncoercive intervention intended to reduce admission rates and resource utilization for patients with acute chest pain. Throughout the intervention, risk information and triage recommendations that had been endorsed by local opinion leaders were communicated without using person-to-person contact. More direct interventions have been more effective [2, 9]. For example, length of stay was reduced by 26% in a recent study [9] in which a utilization reviewer telephoned physicians caring for patients who were at low risk for complications and who remained in the hospital for more than 24 hours. Similarly, admissions to the coronary care unit were decreased by a previous intervention in which risk stratification data were directly provided to physicians by study personnel [2].

There are several possible explanations for the ineffectiveness of our intervention. One is that secular trends toward increased efficiency diminished the chances of any intervention having a significant effect. In this 1990-1992 cohort, only 5% of the low-risk patients who were hospitalized were assigned to the coronary care unit, compared with 51% in a similar 1984-1985 cohort [1]. Another possibility is carryover of the effects of the intervention into control periods.

A more likely cause of the failure of this intervention is the difficulty of changing physicians' behavior by information alone. These data are consistent with previous studies that have shown that education alone or education accompanied by limited feedback may be ineffective in changing physicians' practices [15, 16] and that physicians may not use predictive instruments even when these are readily available [17].

Our findings suggest that different strategies may be needed to provide recommendations to physicians for different types of patients. The goal of our intervention was to influence the management of low-risk patients. However, we recently reported data from a subset of the physicians in this study [18] showing that physicians were more likely to consult the decision aids for patients at high risk for acute myocardial infarction than for patients at low risk. Therefore, the effectiveness of decision aids for low-risk patients may be influenced by physicians' attitudes about when they need assistance.

Our investigation was limited to a single diagnosis at a single institution, and our results do not prove that nonintrusive interventions are always ineffective. At teaching hospitals, decision making is often done among several physicians; thus, an intervention targeted at a single physician may be more effective in a community hospital setting. In addition, our intervention might lead to increased efficiency if applied to a diagnosis or at an institution with greater variability in management patterns.

However, our findings strongly suggest that other strategies will be needed if lowered costs are to be achieved for this important patient population without compromising quality of care. Possible strategies include, but are not limited to, person-to-person interventions such as that described by Weingarten and colleagues [9-11]; a redesign of the process of care, perhaps to include a chest pain emergency facility for the rapid, low-cost evaluation of patients with possible acute myocardial infarction [19]; use of case managers to improve continuity and efficiency of care; and the use of critical pathways to standardize practice patterns [20]. Future research should evaluate both the costs and outcomes of such interventions.