Practice Guidelines and Reminders to Reduce Duration of Hospital Stay for Patients with Chest Pain: An Interventional Trial

Methods

Setting

We did the study at Cedars-Sinai Medical Center, a teaching community hospital that serves West Los Angeles; it has an 18-bed CCU and a 32-bed intermediate care unit. Although a CCU director, cardiology faculty members, cardiology fellows, residents, interns, and medical students care for patients admitted to the CCU, most final hospital discharge decisions are made by private practitioners.

Patients

Men and women older than 30 years who were admitted to the CCU or intermediate care unit with chest pain between 1 September 1991 and 31 August 1992 were eligible for the study. Utilization review coordinators trained to interpret the guideline identified study patients. An earlier version of the guideline was developed by others [20, 21]. We modified the guideline through the consensus of local experts and validated the guideline in several of our previous studies [22, 23, 27-29]. Patients who satisfied any single criterion from the following criteria within 24 hours after hospital admission were not eligible for enrollment in the study: 1) acute myocardial infarction, 2) recurrent chest pain, 3) previous cardiac intervention, 4) a planned cardiac intervention, 5) a previous significant complication, 6) an ongoing intervention, and 7) an unstable comorbid condition. Absence of all seven criteria during the hospital stay defined low risk.

The Intervention

We evaluated the effects of providing to physicians a practice guideline that recommends consideration of early hospital discharge for patients with chest pain who are considered to be at low risk for complications. The study had an alternate-month time series design. During intervention periods, physicians received a structured message posted on patients' charts the day after admission that conveyed risk information and the guideline recommendation (Table 5). Physicians caring for patients at low risk who remained in the hospital more than 24 hours after admission were contacted by a utilization management physician via telephone. A structured interaction message was used to deliver the risk information and guideline recommendation Appendix Table 1. During control periods, no physician contact was made. A 3-day “washout period” corresponding to the last 3 calendar days of each month was maintained to minimize contamination of control patients. There were 12 study periods (6 control and 6 intervention).

All patients who were admitted to the CCU or intermediate care unit with chest pain during the study period were reviewed concurrently by utilization review coordinators (n = 1531) (Table 1). Retrospectively, research nurse reviewers (who were not blinded to the original judgments of the utilization review coordinators) determined the correct risk classification according to the guideline based on complete information available in the medical record. The consensus of the two research nurses served as the “gold standard” for risk determination.

The utilization review coordinator classifications were correct for 88% of patients (Table 1). Of the 13% of patients who were misclassified, 9% of patients were truly not at low risk for complications according to the guideline, yet the utilization review coordinators classified them as being at low risk (yielding false-positive results). Another 4% of patients were at low risk for complications yet they were not concurrently identified by the utilization review coordinators as being at low risk, and thus the intervention was not done (yielding false-negative results). Analyses were done and included all 375 patients with chest pain who were considered to be at low risk and who were admitted to Cedars-Sinai Medical Center during the study period (including patients with true-positive and false-negative results).

Information regarding severity of illness was collected at the time of admission. A time-insensitive predictive instrument served as the study's severity-of-illness measure [5, 30]. Charlson and associates' comorbidity index was used to quantify the number and severity of patients' comorbid diseases [31].

Patients were categorized as having either acute myocardial infarction, unstable angina, noncardiac chest pain, or unknown diagnoses. Acute myocardial infarction was defined as at least one creatine kinase MB enzyme determination of 3.5% or more.

The primary physician was the physician of record; this was usually the physician who cared for the patient before admission. One hundred fifty-five physicians served as the primary physician for the 375 study patients. Frequently, a combination of physicians was involved with the care of individual patients; combinations usually included at least one primary physician (frequently with a covering partner) and often a consulting cardiologist (often with a covering partner for weekends). There were 280 physician combinations caring for the 375 study patients. Because patients were cared for by many physician combinations, individual physicians could not be used as the unit of analysis [32, 33].

Outcomes

Patient instability at discharge was determined for patients admitted to the hospital during control and intervention months using the RAND instability-on-discharge measure [34]. Patients or patients' families were surveyed 1 month after discharge regarding patient survival, hospital readmission, and other problems that might have occurred after hospital discharge. If confirmation of survival could not be obtained from the patient, Equifax (McLean, Virginia), a large computerized database service, was queried to confirm patient survival or death.

Patients were also surveyed about their health perceptions using a summary item from the Medical Outcomes Study short form [35]. In addition, patients were asked to rate the quality of information they received from their physicians at discharge [36].

All cost and charge data were obtained from the hospital's financial database. Charges were based on actual prices charged for services rendered during the hospital stay. Charges were aggregated into routine charges (for example, room rates) and ancillary charges (for example, the price for two acetaminophen tablets). Total costs, which reflected the cost to the hospital for providing a service or good, included indirect and direct costs and were calculated after patient discharge from the hospital [37].

Statistical Analysis

Means are reported ±SD unless otherwise noted. The primary end points for comparison were total length of hospital stay, total hospital costs, and 1-month complication rates. End points were tested using a type I error of 0.05. Statistical analyses were done using the software packages GB-Stat [38], True Epistat [39], SAS [40], or BMDP [41].

One-month complication rates were chosen to avoid framing issues created by earlier hospital discharge. Complications were compared using a chi-square test or Fisher exact test. Confidence intervals around zero numerators were calculated using the method described by Hanley and Lippman-Hand [42].

Continuous data for the study groups were compared using the Student t-test, the Wilcoxon rank-sum test, or both when the data were notably distributed in a non-normal pattern. An adjusted analysis comparing the two study groups with respect to total costs and length of stay was done using a stepwise regression procedure. An initial model was based on patient entry characteristics (patient age, sex, admission source, preadmission residence, initial disposition, comorbidity index, and probability of acute ischemic heart disease). Factors were selected based on a P-to-enter value of 0.10. After the model was finalized, patient groups (control or intervention) were tested for inclusion in the model using a significance level of P ≤ 0.05. The SAS stepwise regression procedure, “STEPWISE,” was used for the analysis [40].

We estimated savings in total costs and length of stay by calculating the difference in these values in the intervention and control groups (including outliers). Because the two study groups were similar and the predictive models had very low coefficients of determination Appendix Table 2, this method provided a good estimate of savings associated with the intervention. We had a second approach as well—calculating expected total costs and length of stay for each person in the intervention group, based on the control group models. For each patient in the intervention group, an estimated cost or length-of-stay savings was measured as the difference between his or her observed and expected values. Ninety-five percent confidence intervals were calculated as the mean ±1.96 times the standard error of the difference distribution.

Results

During the 12-month study period, 1531 patients were admitted to Cedars-Sinai Medical Center with chest pain who were screened for entry into the study. Ninety-nine percent of medical records were available for retrospective analysis. Eighty-six percent of patients were contacted 1 month after discharge and surveyed about their condition. Information on survival status 1 month after hospital discharge was available for 99.7% of patients.

One hundred eighty-three patients with chest pain who were considered to be at low risk for complications were enrolled during intervention periods (1.12 patients per day), and 192 similar patients were enrolled during control periods (1.04 patients per day). The mean patient age was 67.0 ±14.3 years. Fifty-four percent of patients were women; 83% of patients were white, 11% were black, and 6% were of other races.

Sixty-three percent of patients were initially admitted to the intermediate care unit and 37% of patients to the CCU. Forty-four percent of these patients were cared for by a cardiologist as the primary physician, 41% of patients were cared for by a noncardiologist as the primary physician in conjunction with a consulting cardiologist, and 15% of patients were cared for by a noncardiologist without a consulting cardiologist.

Final diagnoses for the study patients were noncardiac chest pain for 22% of patients; unknown diagnosis for 33%; angina or ischemia without infarction for 29%; acute myocardial infarction for 0.5% of patients; and other diagnoses for 15% of patients.

Severity of illness was similar for patients in the intervention and control groups (Table 2). Therefore, no significant differences were noted in physicians' admitting practices during intervention and control months.

Physician Compliance with Guidelines

Physicians complied with the guideline in 69% of cases during intervention periods and in 50% of cases during control periods (P < 0.001). This difference showed that guideline feedback significantly changed compliance with guideline recommendations and physician behavior.

Compliance with the guideline (2-day length of stay for patients with chest pain at low risk) was higher when internists cared for patients (84%) than when cardiologists (58%) or combinations of internists and cardiologists (53%) (P = 0.001) cared for them.

Resource Use Data (Cost and Length of Stay)

The mean length of hospital stay for all study patients was 3.1 ±3.6 days; length of stay in the CCU was 0.45 ±1.45 days, and length of stay in the intermediate care unit was 1.62 ±1.5 days. Use of the guidelines reduced total length of stay by 26% (3.54 ±4.1 compared with 2.63 ±3.0 days; 0.91-day reduction, CI, 0.18 to 1.63; P = 0.02); CCU length of stay was reduced by 47% (0.58 ±1.95 compared with 0.31 ±0.56 days; P = 0.06), and intermediate care unit length of stay was reduced by 28% (1.88 ±1.68 compared with 1.35 ±1.23 days; P < 0.001). After excluding an outlier in the control group, there was still a significant reduction in total length of stay (3.35 ±3.1 compared with 2.63 ±3.0 days; P = 0.02). These analyses were done by comparing patients who were at low risk according to the guideline.

A significant reduction in total costs of $1397 per patient (CI, $176 to $2618; P = 0.03) was achieved during the intervention period. After excluding an outlier in the control group, the difference was still significant ($898; CI, $172 to $1625; P = 0.02). The annual cost reduction was projected to be $550 000 per year.

Had we included all patients who were originally identified by the utilization review coordinators (including false-positive results for patients and excluding false-negative results for patients), a significant reduction in hospital length of stay (3.87 ±4.27 compared with 3.02 ±3.58 days; P = 0.02) and a reduction in total costs of $1232 (P < 0.05) were still achieved for each patient at low risk.

The yearly cost of the intervention was estimated to be $34 400. The total cost savings (direct and indirect costs) afforded by the intervention exceeded the cost of providing the reminder by a factor of 15:1, and the total savings for a 1-year period was approximately $515 000.

Use of cardiac diagnostic tests was examined 1 month after hospital discharge to exclude the possibility that shorter lengths of hospital stay had shifted diagnostic testing from the inpatient to outpatient setting. There was no significant difference in the use of treadmill examinations (17% compared with 12%; P > 0.2) and left-heart catheterizations (2.7% compared with 5.4%; P > 0.2) between patients in the intervention and control periods.

The effect of the intervention in reducing length of stay appeared to persist with time (Figure 1). Moreover, no significant change in physician compliance with the guidelines occurred during the study periods when reminders were discontinued. The control month with the longest lengths of stay was July (when new housestaff began internships).

Outcomes

No patients died during their inclusion in either study group (0%; CI, 0% to 0.8%). No significant differences were found in the number of patients who had cardiac complications, in the number of patients who sustained acute myocardial infarctions, in the types of complications, or in the number of patients who required transfer to a higher level of care between intervention and control groups (Table 3). No significant difference was noted between the instability at discharge (0.08 ±0.29 compared with 0.09 ±0.29 instabilities; P > 0.2) for patients admitted to the hospital during intervention and control periods [34].

No significant difference was found in mortality rates between control and intervention groups when measured 1 month after hospital discharge (Table 4). More patients were readmitted who had been in the control group 1 month after discharge, although no significant difference was noted in the number of patients readmitted because of cardiac conditions.

No significant difference in the overall self-reported health status was noted between the control and intervention groups (Table 4). Also, no significant difference was identified in the number of patients who returned to work, the amount of time before they returned to work, the number who resumed recreational activities, and the amount of time before they resumed recreational activities (Table 4).

Patient Satisfaction

No significant difference was found in patient satisfaction between patients admitted to the hospital during intervention and control periods. A similar percentage of patients would either definitely or probably return to the same hospital (94% compared with 94%; P > 0.2). Most patients in the intervention and control groups thought they stayed in the hospital for an appropriate length of time (77% compared with 76%; P > 0.2).

Statistical Analysis

The regression model for hospital length of stay and total costs is described in Table 6. The mean difference between the observed and expected length of stay in the intervention group was 0.80 ±3.1 days (CI, 0.3 to 1.3 days), with the expected length of stay being greater. The mean difference between the observed and expected total costs in the intervention group was $1168 ±$3463 (CI, $662 to $1674), with the expected costs being greater.

We followed the 9% of patients who were misclassified as being at low risk and suitable for the intervention (see Table 1), yet they were not at low risk according to retrospective classification. We found that none of these patients died during the month after hospital admission. The readmission rate for those misclassified patients was not significantly different from that of control patients [8.9% compared with 9.3%; P > 0.2].

Two research nurses independently and retrospectively reviewed 82 randomly selected records to determine if the patient was at low risk or high risk according to the guideline. The κ value was 0.94 (n = 82).

Discussion

Our study showed that concurrent feedback of practice guideline recommendations significantly changed physician behavior and resulted in changes in patient care. Feedback of the guideline recommendations resulted in patient care that more closely approximated care outlined by the practice guideline, without adding incentives, threatening sanctions, imposing administrative barriers, or adding other compulsory methods to enforce physician acceptance of guideline recommendations. As a result of this intervention, medical care for patients with chest pain who were considered to be at low risk for complications was provided at a lower cost. Moreover, the reduction in health care costs was accomplished without significantly compromising patient outcome or satisfaction.

Our practice guideline recommended early hospital discharge for these patients. Additional prognostic information is available 24 hours after hospital admission that is not available at admission, and this information can improve the accuracy of physicians' triage decisions. Mulley and colleagues [20] and Lee and associates [21] showed that patients with no evidence of acute myocardial infarction or cardiac complications 24 hours after hospital admission derived little benefit from continued care in the CCU. Earlier discharge of patients at low risk appears to be a promising strategy for safely reducing health care costs for such patients.

Our study contributes to understanding practice guideline implementation and evaluation in CCUs and intermediate care units. First, although many practice guidelines have been developed in the past, few studies describe use of guidelines for patient care [28, 43]. The failure of many previously developed practice guidelines to be adopted by physicians is discouraging because the ultimate value of a guideline depends on its effect on patient care [43-46]. Despite the vast resources and effort devoted to guideline development, disproportionately few evaluations of their efficacy, safety, and acceptability when used by health care providers in clinical practice are available [46-48]. Our investigation is one of the first generation of studies to evaluate the use (or nonuse) of guidelines by physicians in clinical practice.

Second, we measured the effect of shortening length of stay on patient outcomes and satisfaction. By appropriately shortening patient length of stay through implementation of the guideline, we maintained excellent patient outcome and satisfaction that were unchanged from when care was provided without using the guideline.

Third, we examined whether decreasing length of stay would shift diagnostic testing from the inpatient to the outpatient setting. Previous studies documented that some cost-containment programs may have “squeezed the balloon” and shifted costs from the inpatient to outpatient setting without a net reduction in overall costs [49]. The cost reduction afforded by our guideline intervention did not appear to be offset by increased outpatient diagnostic testing, although a more rigorous evaluation of outpatient costs after hospital discharge would have been valuable.

Fourth, we measured the efficacy of practice guideline implementation by hospital personnel rather than that of guideline implementation in a research setting [50, 51]. As expected, a quantifiable misclassification rate was found when utilization review coordinators implemented medical practice guidelines, even when the guidelines were explicitly defined, extensively tested, and implemented by experienced utilization review coordinators (see Table 1). The misclassification rate of utilization review coordinators with lesser training and expertise may be greater. This study underscores the concern that the effectiveness of guidelines in actual clinical practice may differ substantially from their efficacy as reported in carefully controlled research environments [29].

Fifth, our study showed the possible failure of education alone to sustain substantial changes in physician practice patterns. When direct feedback of guideline recommendations was withdrawn, physician practice reverted to that observed before initiation of the intervention (see Figure 1). Even after the reminders were provided for almost 1 year, physicians frequently failed to comply with guideline recommendations when they did not receive direct and concurrent reminders. This observation may support the belief that education alone, and education accompanied by limited periods of feedback, may be relatively ineffective in sustaining long-term changes in physician practice [44, 45]. Another possibility is that physicians became accustomed to the discharge reminder when caring for patients with chest pain who were at low risk.

Sixth, after the completion of our study, the Cardiac Advisory and Medical Advisory Committee at Cedars-Sinai Medical Center, which are composed of private practitioners and academic faculty members, approved feedback of the practice guideline recommendations. Feedback of the guideline recommendations has been occurring as part of the operations of the hospital's Utilization Management Department. This extra step of translating research findings to hospital policy would further suggest that practice guideline research could be used by hospitals to control health care costs.

Our study had several important limitations. The guidelines were disseminated at only one community hospital, and therefore the generalizability of our results to other hospitals and health care systems is unknown. The guideline dissemination strategy used in our study was labor-intensive, and not all hospitals have the resources and personnel available to disseminate guidelines in this manner. We found that the cost reduction afforded by the guideline intervention, however, substantially exceeded the cost of the intervention. Also, our study had insufficient power to detect small increases in morbidity, mortality, and readmission rates, as has been the case for many previous studies that investigated less costly strategies of caring for patients with chest pain [10, 11, 13, 18, 20-23, 27].

Another limitation of the study was that there were patients who benefited from hospital care after the second hospital day, and therefore the guideline could not replace careful physician judgment in making discharge decisions. This observation, however, was expected, and it reaffirmed the importance of allowing physicians to over-ride practice guideline recommendations when careful clinical judgment deems this necessary. Because chest pain may be a sign of many different pathophysiologic processes [48], the guideline should be used to complement, rather than replace, physician judgment.

The practice guideline evaluated in this study holds promise for enhancing the efficiency of care and improving health care affordability in the CCU and the intermediate care unit. We found that the practice guideline accompanied by reminders reduced length of stay and health care costs for patients with chest pain judged to be at low risk. Moreover, the reduction in health care costs occurred while maintaining excellent patient outcome, quality of care, and satisfaction. Our results support the future study of the described risk classification strategy and practice guideline at other hospitals and within other health care systems.