Abstract
Background: Much of the research comparing specialists and generalists is from studies of patients who had a myocardial infarction. The present study systematically examined this research.
Methods: Medline was used to search for all articles published from 1990 to 2003 that compared cardiologists and generalists for adjusted mortality rates of patients with myocardial infarction. From each article identified, information was abstracted on factors that could have influenced the comparisons.
Results: The studies consistently found that patients of generalists were at greater risk of mortality from both cardiac and noncardiac risk factors and had higher unadjusted mortality rates. Adjusting for risk factors decreased the differences between cardiologists and generalists. Studies that seemed to do the best job taking into account patient differences had similar adjusted-mortality rates for the cardiologists and generalists. No studies adequately took into account reasons the patient did not have care by a cardiologist, eg, patient preferences, severity of comorbid disease, general health status, or resource availability.
Conclusions: Generalists and cardiologists differ substantially with respect to their patients and practice environments. Results comparing patient outcomes by specialty are often influenced by important patient or resource characteristics that were not taken into account.
In the 1990s, managed care proponents heralded the Medical Outcomes Study as proof that outcomes for patients with chronic diseases cared for by generalists were as good as care delivered by more expensive specialists.1,2 In response to these findings, observational studies were conducted to compare the processes and outcomes of care delivered by specialists and generalists.3,4
Many of these studies were conducted on patients with myocardial infarction. This condition was chosen for research because it has an easily measured and important outcome, mortality, that is influenced by the quality of care. Some studies found that outcomes and processes of care are often better for cardiologists than generalists,3,4 but others found no significant differences.5–7 Although the authors acknowledged the limitations of observational data in principle, some concluded that specialists provide better care to cardiac patients than generalists because of their more focused training, knowledge, and skills.8–10 If this explanation is true, it should have important implications for the organization of the health care system.
To evaluate the evidence for the superiority of specialist care, the present study systematically reviewed articles that compared cardiologists and generalists with respect to their patient mortality rates following a myocardial infarction. Previous articles have not examined how details of variation in adjustment variables and patient selection influence the comparisons of generalists and specialists.
Methods
To identify articles that compared specialists and generalists to outcomes following a myocardial infarction, we conducted a Medline search using the MeSH term Myocardial Infarction. The search was limited to those studies that included at least one of the following terms in the title, abstract, or MeSH subject heading: specialist, cardiologist, generalist, family physician, family practitioner, primary care physician, noncardiologist, or general internist. We eliminated articles that were written before 1990, did not analyze original data, included children, were not written in English, or made no adjustment for confounding when reporting mortality rates comparing generalists and specialists. Only studies of patients in the United States were reviewed because factors influencing the comparison of specialties probably vary by country.
To determine whether additional studies might have been missed by our search, we reviewed all the references in the articles located. We also discussed this work at 2 international meetings with experts in the field.
From each article, 2 research assistants, who were not aware of the purpose of the study, independently abstracted information in 5 domains that have been suggested as important in the evaluation of observational studies: comparability of subjects, exposure or intervention, measurement of outcomes (mortality), statistical analysis, and sponsor or funding sources.11 Although some studies compared process measures relating to the management of the patient, we found no studies that evaluated outcomes other than mortality. Discrepancies in the abstracted data were resolved in discussions between the research assistants and one of the authors (PJ). Taking subject factors into account is especially important in primary care patients because of the heterogeneity of this patient group.12 For this reason, special effort was made to abstract information relevant to confounding that might influence the association of outcome (mortality) and exposure.13
To identify potential weaknesses in study design or analysis, we examined each article from a theoretical framework that included factors potentially able to affect either the selection of health care delivery (generalist or cardiologist) or mortality. This framework includes the structure, process, and outcome model that Donabedian used to describe the quality of care.14 In addition, we added elements from Anderson’s Behavioral Model of Health Services Use that considers environmental factors and patient or population factors as contributing to health care service selection and outcomes.15 A model including these factors is depicted in Figure 1.
Patient factors referred to in the model include those that influence patient risk (socioeconomic status and other demographic variables, cardiac severity, and noncardiac diseases). These factors modify the risk of a bad outcome and, therefore, determine the best treatment and influence patient choices about interventions. Patient factors such as location and preferences for certain interventions influence choices of hospital, physician, and process of care. For example, patients far from urban centers may have delays in care and less access to cardiologists and cardiac resource intensive hospitals.16
Hospital factors have been well documented to contribute to mortality based on several contributors, including size of the hospital, volume of patient admissions with myocardial infarction, type of managed care organization, teaching status, and the presence of revascularization facilities.5,7,17 Physician factors such as board certification, years experience and specialty distribution may also contribute to process of care which in turn influences outcomes.18
The intervention, or exposure, tested in the reviewed studies was physician specialty. The outcome measure used for a given specialty was the odds ratio of mortality for this specialty compared with cardiologists. Most studies reported the adjusted odds ratios comparing mortality between cardiologists and other specialists. One study, however, reported the difference in adjusted mortality rates between cardiologists and other specialties. For this study7 we used a statistical method19 to compute the adjusted mortality rate for each specialty from the overall observed mortality rate and the differences in mortality rate. The adjusted odds ratios were then computed from the adjusted mortality rates.
To test for heterogeneity across studies of mortality rates for a given specialty, we computed a χ2 test for 2 by k contingency tables. We tested the difference between 2 odds ratios using the equation Z = (ln1 − ln2)/√(SE12 + SE22) where Z has a normal distribution with mean zero and variance 1, ln1 and ln2 are the logarithms of the 2 odds ratios, and SE1 and SE2 are the standard errors of these logarithms. Heterogeneity in odds ratios was tested with the Breslow-Day test for homogeneity at the P < .05 level.
Results
Forty-two articles were selected for review; 33 were excluded for the following reasons: 12 were not in the United States, 17 reported no patient mortality results or did not report adjusted mortality rates, 3 were not original research, and one evaluated care in the postmyocardial infarction period. The 11 remaining articles compared mortality rates for generalists to cardiologists before and after adjusting for patient risk factors. Four of these used different components of data from the Cooperative Cardiovascular Project.20,21 Data collected for the systematic review are shown in Table 1.
There was variation in how physician specialty was identified: one study used only board certified physicians,5 one study identified the specialty as the admitting physician,22 and some studies combined noncardiologists into one group.6,7,9,10 Even when studies identified physicians in the same way, eg, attending physician, this can be variously determined by hospitals as admitting physician, discharge physician, or physician who spent the most time with the patient. The percentage of physicians defined in a given way will depend on the hospitals in the data set.
In Table 2 studies are compared for differences between patients of cardiologists and generalists. For most studies, differences between the generalists and cardiologists for a given factor suggested that generalists’ patients were at higher risk. Other risk factors that were not tabulated because they were reported by a few studies, eg, the Atlas severity score9,10,13,18 or the severity score based on factors in the study entitled “Global Utilization of Streptokinase and Tissue Plasminogen Activator for Occluded Coronary Arteries” (GUSTO)5–7,22,23 also showed that the patients of the generalists were at higher risk (see Table 3). Two articles included in our analyses9,13 were not included in this table because they did not show patient risk factor comparisons between specialty groups.
To reduce confounding due to differences in patient risk, all studies reviewed took into account many characteristics that influenced mortality rates after a myocardial infarction. These characteristics included demographic characteristics (age, race, and gender), cardiac risk factors (the 13 GUSTO-1 variables related to cardiac function), comorbidities (eg, elevated creatinine or BUN), functional status (eg, mobility), and hospital characteristics (revascularization facilities and volume of myocardial infarction patients). Differences among the studies are described in Table 4. None of the studies included the reasons why the patient did not receive care from a cardiologist, eg, they were not candidates for invasive procedures either because of their poor health or personal preferences. The studies also did not include variables such as socioeconomic status or the severity of the comorbid conditions (eg, cognitive impairment, chronic obstructive lung disease, or cancer). Only the studies by Ayanian et al6,24 assessed whether cardiologists and generalists collaborated in the care of patients with myocardial infarction.
Unadjusted and adjusted mortality comparisons of generalists and specialists are shown in Table 5. Mortality was assessed at 30 days and 1 year for 5 of the studies and in-hospital for 4 others. There was a statistically significant variation across studies in 30-day mortality rates for patients treated by physicians in a given specialty. There was also a significant variation in the unadjusted odds ratios comparing generalists to cardiologists. For all studies the unadjusted mortality rates were lower for cardiologists than for other specialties. For comparisons of family physicians, internists or combinations of generalists to cardiologists unadjusted odds ratios for in-hospital or 30-day mortality ranged from 1.14 to 1.59 and unadjusted 1-year mortality rates ranged from 1.37 to 1.66. Because the generalists treated higher risk patients in all the studies, however, the odds ratios were reduced, sometimes to less than 1.00, after adjusting for risk factors.
We examined whether the quality of the risk-adjustment was associated with results. Studies were considered to have better risk-adjustment if the cardiologists and generalists had similar patients ie, the patients were similar age6 or all patients were less than 659 or the study adjusted for a comprehensive set of risk factors including patient comorbidities and functional status.5 These studies all had adjusted odds ratios near 1.00. The studies with the greatest differences between generalists and specialists in patient age10,18 for subjects older than 659 or that did not adjust for comorbidities22 found that generalists had higher mortality than cardiologists, ie, the odds ratios were significantly >1.
In Table 5 we did not include a study of patients who survived a myocardial infarction for at least 3 months.24 In that study the mortality rate in the postmyocardial infarction period for patients managed by generalists alone was 19.1% compared with 11.8% for patients managed by cardiologists alone or with a generalist, OR = 1.75, P < .001. After reducing confounding by using propensity scores to select patients that were similar with respect to 36 variables representing clinical characteristics, hospital care, medications at discharge and hospital characteristics, the mortality rates were 18.3% for the generalists and 14.6% for the cardiologists, OR = 1.32, P < .001. Another finding in this study was that patients of cardiologists were much more likely than patients of generalists to have angiography (26.8% vs 16.7%), angioplasty (11.8% vs 6.9%), bypass graft surgery (11.9% vs 7.0%), cardiac rehabilitation (36.4% vs 29.0%), or exercise stress testing (61.4% vs 52.8%). As with the other studies, this study did not take into account the reasons why the patient did not receive care from a cardiologist.
Discussion
We systematically reviewed studies that compared cardiologist and generalists for outcomes of patients with myocardial infarction. The studies consistently found that patients of generalists were at greater risk of mortality from both cardiac and noncardiac risk factors and had higher unadjusted mortality rates. Studies varied as to whether or not generalists had higher mortality rates after adjusting for risk factors.
The generalists may have treated higher risk patients because patients who were too old or sick to want heroic intervention measures or revascularization procedures did not want to be referred to cardiologists. It is also possible that generalists’ patients were at higher risk because they had greater delay in accessing care. It is unclear why adjusting for patient risk factors did not give consistent findings across studies. We hypothesized that the procedures used to take into account patient risk were not adequate and that the most valid results would be those that had the least differences between the patients of generalist and specialists or that did the best job of accounting for differences in patient risk. The best studies may have been the ones that had similar age patients for generalists and specialists,6 that excluded older patients,9 or that adjusted for the most complete set of risk factors.5 All these found that cardiologists and generalists had similar adjusted mortality rates. On the other hand, the studies that had the greatest patient differences between specialties (as indicated by large differences in age or baseline 1-year mortality rates),10,18,23 or that used fewer risk factors for adjustment22 found the greatest differences between the cardiologists and generalists.
No analyses took into account the reason cardiologists were not consulted. These reasons may include patient location, socioeconomic factors,25 severity of comorbidities, or patient-perceived health status. Results from a study that examined patient refusal of cardiac revascularization suggest that patient preference is also a reason.26 That study found greater refusal among patients who were older, female, minorities, in smaller hospitals, or cared for by generalists.
Another contributor to differences between specialties is the method used to handle transfers. Transfers involved a high percentage of patients in most studies reviewed (see Table 1). It has been shown that transferred patients have a much lower in-hospital mortality rate than other patients (4.6% vs 11.4%).27 For the studies reviewed, transferred patients were either eliminated5,7,9,10,23 or assigned to a cardiologist22 or shared care.6 Both these methods of handling transfers assign lower risk patients to groups that involve care by a cardiologist. Primary care physicians will be left with high-risk patients who are not transferred. Including a variable for patient transfer in a regression equation will not be an adequate adjustment because generalist patients who are not transferred will be at higher risk than cardiologist patients who are not transferred. The best way of analyzing transferred patients is to associate their outcome to the physician who transferred them.
This review is not the first to suggest that patient and hospital factors may contribute to different outcomes for cardiologists and generalists.5,7 However, this suggestion has not been proven or generally accepted,28 and even the need for careful adjustment of confounding variables has not been universally recognized.29
Although this review did not resolve the appropriate role for generalists in managing patients with myocardial infarction, it does suggest what information should be collected for future studies. In particular, it would be helpful if generalists documented when care deviates from established standards by noting the contributions of patient preferences, environmental limitations, and unique patient circumstances. Studies could also improve their methods for accounting for patient risk (clinical, comorbid, and demographic), access of health care resources, and transfer to another facility. To the extent possible, study information should be collected by abstracters who do not know the purpose of the study or the specialty of the involved physicians.
The lower patient mortality rates following a myocardial infarction of cardiologists than generalist have led some authors to conclude that cardiologists know better than generalists how to manage myocardial infarction.8,9 The self-evident correctness of the conclusion may have obscured the methodological deficiencies in the studies that support it. However, the conclusion results from a misunderstanding of the study design, which did not really evaluate the relative expertise of physician specialties. Because generalists can usually refer to a cardiologist, the studies actually evaluated whether generalists made timely and appropriate referrals. If they did not, then perhaps patients who may be having a myocardial infarction should go directly to a cardiologist. Because taking responsibility for care from generalists is a major policy decision, it should only be made with substantial evidence. Evidence from the studies reviewed in this manuscript was questionable. In all the studies, generalists treated higher risk patients, and some of these patients may have had needs other than survival. To adequately assess the influence of physician specialty on patient outcome, studies should carefully consider patient needs, patient risk, and the realistic medical options available.
Acknowledgments
We thank Mary Charlton, Dedra Diehl, Kevin Lynch, and Laurie Wallace for technical support of this work.
Notes
Funding: This work was supported in part by R01 HS10739-01 from the Agency for Healthcare Research and Quality.
Conflict of interest: none declared.
Prior presentation: “An assessment of the validity of observational studies to infer specialty differences in cardiac quality of care” Arthur Hartz and Paul James, North American Primary Care Research Group, Banff, British Columbia, 2003.
- Received for publication July 6, 2005.
- Revision received November 11, 2005.
- Accepted for publication November 22, 2005.