EVIDENCE

This chapter evaluates the evidence in the literature around four questions: (1) Are outcomes better in teaching hospitals versus nonteaching hospitals? (2) Is there a difference in outcomes between for-profit and not-for-profit hospitals? (3) Is there a difference in outcome between urban and rural hospitals? (4) How do we interpret the evidence for patients undergoing orthopedic procedures within the context of the first three questions?

Several issues must be considered to resolve these questions, with the major one being the considerable heterogeneity of studies with respect to diseases and procedures, risk adjustment, types of data, and settings.^2,5

Studies report outcomes for patients with a variety of conditions; however, most studies are on cardiovascular disease, mainly congestive heart failure, acute myocardial infarction (AMI), or stroke.^3,4,6–9 Only five studies included patients undergoing orthopedic procedures, and these were in patients who had received surgery for total joint replacement¹⁰ or hip fractures.^{4,7, 11, 12} No studies of outcome in patients with fractures as a result of major trauma have been reported.

No randomized, controlled trials have been identified, and it is unlikely that this will happen in the future. All studies have been observational because this is the most feasible design to address these questions.¹ Although this level of evidence may be considered weak, a well-conducted observational study can contribute valuable information. Many studies are retrospective cohort studies, a relatively robust observational design, not as subject to bias as some other study designs, and these have an important part to play in medical research.¹³ The main concern with cohort studies is residual confounding by unmeasured variables related to the patients and the outcome. This is problematic with the use of administrative data, because usually only demographic information and some measures of comorbidities are available for risk adjustment in the analyses.

Nearly all of the evidence available is based on either clinical chart reviews or administrative data.² Because administrative data are collected for billing, rather than research purposes, problems with coding are an issue.¹⁴ These problems may differ among databases. Nevertheless, although those such as the Health Care Financing Administration data, which has been used in many U.S. studies,^{7,10, 15} and the Canadian Institutes of Health Information data, which has been used in many Canadian studies,^{11,16, 17} have been validated,^14,18 to some extent, this remains a problem inherent in all studies. The administrative data in the U.S. studies are mainly for Medicare patients,^4,10 and this could introduce some bias, if these patients are older and sicker than the average patient. This is less of a problem in countries with universal healthcare, such as Canada, where administrative databases contain information on all patients.

Another challenge in evaluating the evidence is that most of the studies that examine these issues have been done in the United States, where the hospital system is more complex, compared with Canada, for example. In the United States, all teaching hospitals are not-for-profit; however, nonteaching hospitals can be either for-profit or not-for-profit. Thus, the issue of whether teaching hospitals are different from nonteaching hospitals may be confounded by their funding status (for-profit or not for-profit). For example, some studies have compared not-for-profit teaching hospitals with not-for-profit nonteaching hospitals,¹¹ whereas others have compared not-for-profit teaching hospitals with for-profit nonteaching hospitals.⁴ This could lead to inconsistent conclusions.

Two recent meta-analyses have attempted to resolve the questions of whether outcomes are better in teaching hospitals versus nonteaching hospitals² and whether mortality rates differ between private for-profit and private not-for-profit hospitals.¹

The first analysis identified 132 eligible studies that included 93 on mortality and 61 on other outcomes.² Twenty-two of these assessed both. A wide range of patient groups were represented by these studies; most were patients with cardiovascular disease, and only two involved orthopedic patients. Most studies were done in U.S. hospitals. Ten studies were done in Canada, 9 in the United Kingdom, 4 in Norway, and 12 in other countries. Using all adjusted mortality data, the authors report a relative risk (RR) for teaching hospitals of 0.96 (95% confidence interval [CI], 0.93–1.00). Subgroup analyses involving data sources (clinical or administrative) and location of study (United States or other) yielded no important differences. The authors report some evidence that teaching hospitals performed better for certain diagnoses (breast cancer, cerebrovascular accidents, and mixed diagnoses). Because of the small numbers of studies, no information could be calculated for orthopedic procedures. This was an extensive and thorough meta-analysis that suggests that differences between teaching and nonteaching hospitals are minimal.

The second meta-analysis by Devereaux and colleagues¹ compares mortality rates between private for-profit and private not-for-profit hospitals. Fifteen studies were included in this analysis. In an attempt to remove hospital teaching status as a confounding influence, they included results from for-profit nonteaching and not-for-profit teaching hospitals when they were available. They also avoided adjusting for variables under the control of hospital administrators that could be influenced by a profit motive and affect mortality, such as nursing staffing levels. Three of the studies in their analysis included orthopedic patients (hip fractures, lower extremity fractures, hip replacement).^{7,10, 15} One of the three included all patients in the dataset but did not differentiate diagnoses in the analysis.¹⁵