Quality of Care for Patients Hospitalized With Heart Failure: Assessing the Impact of Hospitalists

Background  The quality of care provided to patients hospitalized for heart failure has been shown to vary by physician, hospital, and region. Hospitalists appear to reduce costs and length of stay, yet their impact on quality of care is less certain.

Objective  To compare quality of care and resource utilization among patients with heart failure treated by hospitalists and nonhospitalist general internists.

Methods  We reviewed the medical records of patients with a principal diagnosis of heart failure between April 1, 1999, and March 30, 2000, at a 550-bed community-based teaching hospital in Massachusetts. We evaluated quality of care by measuring adherence to a set of commonly used process measures and compared resource utilization using severity-adjusted length of stay and costs.

Results  The analysis included 280 patients, accounting for 326 heart failure admissions: 20 hospitalists cared for 137 (42%) cases, while 65 nonhospitalists cared for 189 (58%). Of 137 hospitalist cases, 129 (94%) had new or prior left ventricular ejection fraction testing results documented during the hospitalization compared with 165 (87%) of 189 nonhospitalist cases (P = .04). In cohorts of ideal candidates, performance rates for hospitalist and nonhospitalist cases were similar for prescriptions of angiotensin-converting enzyme inhibitors or angiotensin receptor blockers for patients with ejection fractions lower than 40% (97% vs 96%; P>.99) and warfarin for patients with atrial fibrillation (60% vs 55%; P = .64). Rates of comprehensive discharge counseling was similar in the 2 groups. Multivariable modeling did not substantially alter these findings. After adjusting for differences in severity, patients treated by hospitalists had a shorter length of stay but similar overall costs when compared with those treated by nonhospitalists.

Conclusion  Compared with nonhospitalists, hospitalists were more likely to document assessment of left ventricular function and their patients had a shorter length of stay.

HEART FAILURE affects some 5 million Americans¹ and results in approximately 200 000 deaths each year. It is the leading cause of hospital admission among Medicare beneficiaries, with annual costs estimated to exceed $20 billion. Measured against standards developed from clinical practice guidelines from the Agency for Healthcare Research and Quality, the American Heart Association (AHA), and the Advisory Council to Improve Outcomes Nationwide in Heart Failure,^1-3 the quality of care for patients with heart failure has been shown to vary by physician, hospital, and region.^4-7 In light of these findings, improving the quality of care for patients hospitalized with heart failure has been a high priority for the Health Care Financing Administration's (HCFA's) Medicare program as well as the Joint Commission on the Accreditation of Healthcare Organizations (JCAHO).^8,9 Hospitalists are physicians whose primary professional focus is the general medical care of hospitalized patients.¹⁰ In settings ranging from large academic medical centers to small community-based hospitals, hospitalists are playing an expanding role in providing care for hospitalized children and adults.^11-15 A recent workforce projection predicted that the field may ultimately grow to as many as 10 000 to 30 000 and that hospitalists may eventually come to dominate the inpatient care arena.¹⁶ As a result of the enormous financial pressures facing many American hospitals, much of the early interest in hospitalists has focused on their potential role in reducing length of stay and lowering costs.^13,17 There has been less attention paid to the effect of hospitalists on quality of care, and evidence addressing this topic is limited.¹⁸ Because the growth of the hospitalist model carries major implications for the care of patients with chronic illnesses such as heart failure, we sought to determine whether the quality or costs of care for patients with heart failure, as measured by a variety of process and outcome measures, differed between hospitalists and their nonhospitalist generalist colleagues.

We conducted a retrospective review of medical records at Baystate Medical Center, a 570-bed community-based teaching hospital in Springfield, Mass, that serves as the western campus of the Tufts University School of Medicine. Three unique hospitalist groups (1 large academic faculty practice and 2 smaller private group practices) provide care for patients admitted to the hospital's general medical and cardiology units. As well as caring for patients from the teaching clinics, faculty hospitalists are responsible for caring for all patients with no identified primary care physician. In addition, numerous small group practices continue to care for their own patients who require hospitalization. Permission to carry out the study was obtained by the institutional review board at Baystate Medical Center.

We identified consecutive patients discharged with a principal diagnosis of heart failure (International Classification of Diseases, Ninth Revision, Clinical Modification codes 428-428.1, 428.9, 402.01, 402.11, 402.91, 404.01, 404.11, 404.91) for a 12-month period between April 1, 1991, and March 30, 2000. An experienced nurse abstractor reviewed the medical records of potential cases. Patients were excluded from the study if the diagnosis of heart failure could not be validated from a review of the medical record. The coding of heart failure was considered correct if the patient had symptoms, signs, or radiographic abnormalities consistent with a diagnosis of heart failure. Patients were also excluded if the attending of record was a cardiologist or other medical subspecialist, if they were transferred to another acute care institution, or if they were designated as receiving "comfort measures" only. Using a previously published definition, physicians who spent at least 25% of their time caring for the hospitalized patients of other physicians were considered hospitalists.¹⁹

The measures of quality used in this study were derived from a set of indicators developed by the JCAHO for use in their "Core Measures Initiative."⁸ The measures were produced by the JCAHO through a consensus-based process that solicited in-depth input from hospitals, health care purchasers, consumer groups, performance measurement systems, state medical societies, and professional organizations. The 5 measures for heart failure were selected for the ease with which they could be defined and measured, their likelihood for improving health outcomes, and their reliability and validity. Each indicator was evaluated in a restricted sample of patients who were considered "ideal" candidates for the intervention (eg, for the smoking cessation counseling indicator, only patients who were active smokers were included in the denominator). The quality indicators for patients with heart failure included the following:

• Measurement of left ventricular ejection fraction before or during admission

• Use of angiotensin-converting enzyme (ACE) inhibitor or angiotensin receptor blocker (ARB) at discharge for patients with left ventricular ejection fraction lower than 40% and no contraindications to its use

• Prescription of warfarin at the time of discharge for eligible patients with atrial fibrillation and no contraindication to its use

• Provision of smoking cessation advice or counseling during hospitalization for patients with a history of smoking within the past year

• Documentation regarding discharge medications, daily weight monitoring, the importance of diet, activity level, and procedures to follow if symptoms worsen.

From previous studies,^6,20 we estimated ACE inhibitor or ARB use rates in ideal candidates to be approximately 70%. The 1-year period selected for this study was intended to provide enough cases to have an 80% power to detect a 15% difference in ACE or ARB use rates between hospitalist- and nonhospitalist-treated cases with an α of .05.

In addition to collecting information required to measure adherence to the various quality measures, we recorded patient age, sex, ethnicity, language spoken, whether the patient was a resident of a nursing home, whether they had been transferred from another acute care facility, whether house staff coverage was provided for the patient, and whether a cardiology consultation was obtained during the hospitalization. A data collection tool was developed and pilot tested to abstract relevant data from medical records using TELEform (standard version 6.2; Cardiff Software Inc, Vista, Calif) and data were then entered into a computerized database directly via facsimile. Abstraction errors were reduced by providing detailed data definitions and by a reabstraction of a random sample of medical records.

Summary statistics for the overall sample were constructed using simple frequencies and proportions for categorical data and means and SDs for continuous variables. We assessed the association between potential confounders and both the physician type and each quality of care indicator using χ² tests for independence. Factors that were found to be associated with both physician type and a particular quality of care indicator were considered confounders. Multiple logistic regression models were used to examine the relationship between physician type and each of the dichotomous quality variables. For each quality of care indicator, a model was constructed including the physician type and identified confounders. The significance of each coefficient in the model was evaluated, and nonsignificant variables (P≥.15) were eliminated from the model one at a time, starting with the variable with the largest P value. Interaction terms were also evaluated using similar criteria. When all statistically nonsignificant terms were eliminated from the model, overall fit was assessed using the Hosmer-Lemeshow Goodness of Fit Test. All analyses were carried out using SAS (version 8.0, SAS Institute Inc, Cary, NC). Mantel-Haenszel–adjusted relative risks for hospitalists relative to nonhospitalists for performance on quality of care indicators were computed for the final models.

Cost and length of stay data were obtained directly from the hospital's cost accounting system (Health Management Systems, El Segundo, Calif), and figures represent actual costs and not charges. Comparisons of length of stay and hospital costs between hospitalist and nonhospitalist cases were made using the All Patient Refined–Diagnosis Related Groups (APR-DRG) patient classification system (Version 15.0; 3M Corporation, Minneapolis, Minn) to adjust for differences in severity, which is a risk adjustment method that uses secondary diagnoses to assign patients to 1 of 4 levels of severity (mild, moderate, major, and severe). Severity adjustment using the APR-DRG system was limited to patients assigned to APR-DRG 127 (heart failure and shock). Owing to the nonnormal distributions of length of stay and cost per case, we used Kruskal-Wallis analysis of variance (adjusting for severity) to evaluate differences between hospitalist and nonhospitalist cases.

The analysis included 282 patients, accounting for 326 heart failure admissions; 20 hospitalists cared for 137 cases, while 65 nonhospitalists cared for the remaining 189. Each hospitalist cared for a median of 5 heart failure cases, while each nonhospitalist cared for a median of 2 cases (P<.001). Regarding the patients in this study, the mean age was 74 years, 57% were women, most (62%) were white, English was the most (91%) common language spoken at home, 13% were residents of nursing homes, and 2% had been transferred from another acute care facility. Overall, the average length of stay was 5 days and the in-hospital mortality rate was 5%.

Hospitalists, compared with their nonhospitalist colleagues (Table 1), cared for a larger proportion of male patients (50% vs 37%; P = .02), and their cases were more likely to receive house staff coverage (61% vs 29%; P<.001). Patient characteristics including age, race, language spoken at home, whether the patient had been transferred from another acute care facility, whether the patient was a resident of a nursing home, and rates of cardiology consultation were similar in the 2 groups.

Of hospitalist cases, 94% had the results of new or prior left ventricular function testing documented during the hospitalization compared with 87% of nonhospitalist cases (P = .04) (Table 2). Among ideal candidates, performance rates for ACE inhibitor or ARB use for patients with ejection fractions below 40% and warfarin prescription at the time of discharge for ideal patients with atrial fibrillation were similar in the 2 physician groups. Both physician groups performed poorly with respect to smoking cessation. Rates of counseling at the time of discharge regarding medication use, crisis management, diet, and weight monitoring varied, but levels were similar for physician groups for each type of counseling. The mortality rate was 5.8% for hospitalist-treated cases and 4.8% for nonhospitalist-treated cases (P = .66). Fourteen patients (10%) cared for by hospitalists were readmitted within 30 days of discharge compared with 26 (14%) of nonhospitalist-treated cases (P = .31). Overall, one half of the patients cared for by hospitalists were discharged within 3 days of admission, whereas half of the nonhospitalist cases were still hospitalized on day 4. The median cost per case for patients cared for by hospitalists was $3859 compared with $3932 for those cared for by nonhospitalist generalists.

Age, sex, race, nursing home residency, transfer status, house staff coverage, and cardiology consultation were each associated with specific quality of care indicators (data not shown). Adjusting for the confounding effects of sex on the assessment or documentation of left ventricular function had minimal effect on the overall point estimate but did increase the size of the 95% confidence interval (Table 3). Moreover, this process did not reveal any significant differences in performance on other quality of care indicators between hospitalists and nonhospitalists (Table 3).

Of the 326 study patients, 292 (90%) were assigned to APR-DRG 127 and included in this analysis. Using the APR-DRG method of severity adjustment, most (87%) of the cases was classified as either moderate or major severity. Patients treated by hospitalists had an overall higher level of severity (Table 4), with 45% being classified as major or severe compared with 29% for nonhospitalists. After adjusting for this difference in severity, cases managed by hospitalists had a shorter length of stay than those managed by nonhospitalists (P = .03). This effect was most pronounced among the 88 patients in the major severity category, wherein cases managed by hospitalists had a median length of stay that was 2 days shorter than those managed by nonhospitalist generalists. Although costs varied across each severity stratum, there were no overall significant differences between the 2 groups (Table 4).

As one of the leading indications for hospitalization nationwide, the care of patients with heart failure has been the subject of a great deal of attention by both governmental and professional organizations. Well-documented variations in practice^6,21 have led to a proliferation of clinical practice guidelines and quality improvement initiatives aimed at improving quality of care while reducing costs. Coinciding with these efforts, the growth of the hospitalist model has dramatically altered the landscape of inpatient care for patients with chronic medical conditions, such as heart failure, and may have important implications for those interested in addressing quality of care.

The present study was designed to compare the quality and resource utilization patterns for patients with congestive heart failure between those who were cared for by hospitalists and those cared for by nonhospitalist general internists. Unlike prior studies of hospitalists, in addition to comparing costs and length of stay, we attempted to measure quality of care by focusing on a series of evidence-based process measures that are believed to result in improved health outcomes. After adjusting for the confounding effects of age, sex, and house staff coverage, we found that hospitalists were somewhat more likely to assess or document left ventricular function, while the use of ACE inhibitors or ARBs, use of warfarin for patients with atrial fibrillation, and comprehensive patient counseling at the time of discharge were similar. Although assessment of left ventricular function and use of ACE inhibitors and ARBs were generally high, like others⁶ we noted important opportunities to improve practices surrounding warfarin use in eligible patients and to improve patient education and counseling. We found small but clinically significant differences in length of stay between hospitalists and nonhospitalist physicians that persisted even after adjusting for differences in severity of illness.

A number of previous studies have examined the quality and outcomes of care for patients hospitalized for treatment of heart failure using similar process measures.^4-7,22 These studies have demonstrated differences in the performance of individual states, of hospitals within and across states, and of generalist and specialist physicians. In addition, several studies have compared outcomes between hospitalist and nonhospitalist physicians, although none has focused on a single condition such as congestive heart failure or has attempted to measure quality of care using widely accepted process measures. Wachter et al¹⁷ demonstrated reductions in length of stay and costs on a managed care service that was staffed by hospitalist attendings. In that study, quality of care was assessed using a health status questionnaire coupled with readmission rates and mortality. While these measures provide valuable information, they are often insensitive to important differences in quality and do not offer insight into how outcomes may be affected by specific physician practice. In similar fashion, Meltzer et al²³ compared the outcomes at an academic medical center and found that hospitalists had lower lengths of stay, costs, and mortality compared with their nonhospitalist counterparts. We are not familiar with any previous studies that have examined differences in the quality of care between hospitalists and their generalist colleagues focusing on evidence-based process measures. Advocates argue that the evaluation of process measures provides information that can more readily be translated into changes in practice that can improve patient outcomes. In other words, for those interested in improving quality of care, process measures are more directly "actionable" than outcome measures.

There are several significant limitations to this study. First, because the study was carried out at a single institution, it is difficult to know how well the small differences we observed in quality of care and length of stay can be considered to be indicative of hospitalist practice generally. However, despite the study's single hospital design, we included a heterogeneous array of hospitalists practicing in 3 distinct physician groups: 1 academic faculty practice and 2 private group practices. Second, we used a previously described definition of a hospitalist as a physician who spends 25% or more of his or her time caring for the hospitalized patients of other physicians. As a result, the physicians that we categorized as hospitalists demonstrated significant variation in the amount of time they spent in the hospitalist role from 25% for physicians in the academic faculty practice to 100% in the private group practices. If physician performance is directly related to the proportion of time spent in the hospitalist role, then our study may have been less likely to demonstrate differences in quality than if we had only included full-time hospitalists. On the other hand, because the hospitalist physicians in our study were representative of the diversity of hospitalist practices nationwide,²⁴ our findings may be more generalizable than studies focused on only academic or private practice models. A further limitation of the study relates to the selection of an appropriate comparison group of physicians. Because hospitalists care for more than half of all medical admissions at our institution, the primary care physicians who continue to provide inpatient care may be better or worse than primary care physicians elsewhere. Also, our institution has developed and uses a practice guideline and accompanying order set for exacerbations of heart failure. To the extent that these quality improvement interventions reduce variation in practice, we are less likely to note differences between the 2 physician groups.

Because of the study's retrospective design, it is difficult to know whether performance rates on measures related to counseling and education were due to documentation problems, implementation problems, or both. Moreover, simply carrying out and documenting such measures as smoking cessation counseling does not ensure that the counseling was done well. In addition, some of the quality measures we chose may be less reflective of individual physician practice and more indicative of institutional or nursing practice. This may explain why we observed very little difference in rates of patient counseling at the time of discharge between the 2 groups. Nevertheless, the JCAHO and the HCFA have chosen to incorporate these measures into ongoing national quality improvement initiatives.

We used the APR-DRG system for severity adjustment. This system uses diagnosis codes to measure comorbid conditions and extent of disease.²⁵ The APR-DRG results may be biased by random or systematic errors in coding. In addition, the APR-DRG system does not incorporate physiologic measures (eg, actual ejection fraction and blood pressure) to determine severity. Unmeasured severity of illness may account for some of the differences between the hospitalist- and nonhospitalist-treated cases of heart failure. Nevertheless, APR-DRG is a widely used severity adjustment method that was associated with both hospital lengths of stay and costs at our own institution.

In conclusion, while demonstrating lower length of stay, hospitalists were somewhat more likely than nonhospitalist generalists to assess or document the results of left ventricular testing while caring for patients hospitalized with heart failure. Further studies should be carried out to better evaluate the impact of hospitalists on quality of care for patients with heart failure and other conditions that necessitate inpatient treatment.

Accepted for publication October 15, 2001.

This study was presented at the annual meeting of the National Association of Inpatient Physicians, Atlanta Ga, March 28, 2001.

Corresponding author: Peter K. Lindenauer, MD, MSc, Division of Healthcare Quality, Baystate Medical Center, 759 Chestnut St P-5928, Springfield, MA 01199 (e-mail: Peter.Lindenauer@bhs.org).