What Role Do Neurologists Play in Determining the Costs and Outcomes of Stroke Patients?
Background and Purpose Despite growing concern over the large numbers of specialists in the United States, little information is available on how stroke treatment varies by the specialty of the attending physician. This study compares the costs and outcomes of acute stroke patients by physician specialty, especially between neurologists and other specialists.
Methods We selected a random sample of Medicare patients aged 65 years and older admitted with cerebral infarction between January 1 and September 30, 1991, identified from the principal diagnosis on Medicare Provider Analysis and Review records. All Medicare claims for these patients were extracted from the date of admission through 90 days. The attending physician was identified as that physician billing for routine hospital visits during the first 7 days of the stay.
Results Neurologists treating stroke patients were significantly more expensive than other physicians but obtained better outcomes. Ninety-day mortality rates for patients treated by neurologists were significantly lower than those for other specialists. These cost and outcome differences persisted even after adjustment for patient age, comorbidity, hospital teaching status, and other characteristics. Compared with other attending physicians, neurologists were significantly more likely to order diagnostic cerebrovascular tests (especially brain MRI scans), more likely to prescribe warfarin, and more likely to discharge patients to inpatient rehabilitation facilities.
Conclusions Systematic triaging to neurologists based on clinical characteristics unmeasured by administrative data might explain these observed differences between neurologists and other physicians. Alternatively, these specialists may have been better able to identify the mechanism of stroke, information that then affected the course of treatment. Given current pressures to substitute generalists for specialists, however, more research is needed on these stroke treatment differences.
A growing number of national organizations have expressed concern over the relatively large numbers of specialist physicians in the United States.1 The expensive “high-tech” tests and procedures favored by many specialists have been cited as a major factor in healthcare cost inflation.2 Subspecialists (such as cardiologists) have been found to be significantly higher users of tests and procedures, even after adjustment for patient mix.3 Much of the concern over specialization has focused on surgeons and those medical subspecialists, such as cardiologists and gastroenterologists, who perform invasive diagnostic procedures. Medicare physician expenditures for acute myocardial infarction increased more than 50% from 1985 to 1989, for example, largely because of the growing involvement of cardiologists and the associated increase in service intensity, especially cardiac catheterization.4 At the same time, a recent study found that cardiologists were more knowledgeable about the survival benefits of alternative drug therapies for acute myocardial infarction than internists and FPs.5 Furthermore, they were more likely to prescribe those drugs whose benefits have been proved in randomized clinical trials, eg, thrombolytic agents.
By contrast, there has been little, if any, discussion of neurologists. In part, this may be due to their relatively small numbers; there are only 3.25 non–federal patient care neurologists per 100 000 population compared with 5.65 cardiologists and 38.25 internists.6 It may also reflect the more limited number of expensive tests and procedures performed by neurologists compared with other specialties. Because neurologists are the specialists with the greatest training in cerebrovascular disease, the third most common reason for hospitalization among the elderly,7 we wanted to compare their treatment of acute stroke patients with that of other specialists.
A small number of controlled trials (all conducted in Europe) have found superior outcomes for stroke patients managed by neurologists. Stroke patients treated in stroke units were more likely to be discharged home and to achieve better functional outcomes than those treated in general medical wards.8 9 Although there are no standardized definitions of stroke units, commonly shared features are multidisciplinary team care, early rehabilitation, and early involvement of a neurologist. Recently, a Finnish trial randomized elderly stroke patients to the departments of medicine and neurology and found better outcomes and lower costs for those patients treated by neurologists.10 Although no randomized controlled trials have been conducted in the United States, a small cohort study in North Carolina also found that stroke patients treated on neurology services experienced significantly lower mortality rates than stroke patients treated on other services.11
We sought to compare the costs and outcomes of acute stroke patients treated by neurologists with those of stroke patients treated by other physician specialties, using Medicare claims data. Since neurologists are disproportionately clustered in teaching hospitals, we also sought to separate the effects of teaching status from those of physician specialty. (Almost one third of all neurologists [29%] reported that they had a full-time academic appointment in 1991 to 1992, and another 34% reported a clinical [part-time] appointment.12 )
Subjects and Methods
Sample and Data
We limited our study of acute stroke patients to those admitted with cerebral infarction (and not hemorrhagic stroke), since the two groups vary dramatically in both costs and outcome. Hemorrhagic stroke patients are more expensive to treat and more likely to die. Furthermore, cerebral infarctions account for the majority (89%) of stroke admissions (unpublished Medicare data from the Stroke PORT, 1991). Cerebral infarction patients were defined as those admitted with principal diagnoses of occlusion of cerebral arteries (ICD-9 code 434) or acute but ill-defined cerebrovascular disease (ICD-9 code 436). We did not include patients with a principal diagnosis of occlusion and stenosis of precerebral arteries because medical record validation of discharge abstracts has shown that the vast majority of events with this ICD-9 code (433) were actually stroke sequelae or nonstroke.13
We selected a random 20% sample of Medicare patients aged 65 years and older admitted with cerebral infarction (hereafter referred to simply as stroke) between January 1, 1991, and September 30, 1991 (n=38 612). Admissions were identified from the principal diagnoses on Medicare Provider Analysis and Review records. All Medicare claims for these patients were then extracted from the date of admission through 90 days; these included records for all acute hospital stays, long-term hospital (primarily rehabilitation) stays, SNFs, hospital outpatient departments, home health care, and physician/supplier services. We also extracted any acute hospital records during the preceding 3 years with a principal diagnosis of cerebral infarction, TIA, or hemorrhagic stroke. Data on patient characteristics and date of death (if applicable) were obtained from the 1991 denominator file of the Health Care Financing Administration.
Identifying Attending Physicians and Teaching Hospitals
We identified the attending physician as that physician billing for routine hospital visits (based on Current Procedural Terminology–4 codes on physician claims) during the first 7 days of the hospital stay. Seven days were selected, rather than the entire admission, to avoid misclassification of patients whose care was transferred to another type of specialist after the acute event had passed. Neurologists were compared with internists, FPs, and “all other” specialties. The latter was a heterogeneous category of specialties but largely constituted medical subspecialists, especially cardiologists. For a large number of patients (24.7%), both a neurologist and a primary care physician (either internist or FP) were billing for routine hospital visits. We examined these “combination” cases as a separate category.
Although some error may have been introduced through this approach of identifying attending physicians, it is the only possible method with claims data. There is no reason to believe that any error would be systematically correlated with either costs or outcomes. Previous research has shown that this approach has considerable face validity.4
Because teaching hospitals are heterogeneous with respect to scope of services and other factors that may influence stroke management, we classified hospitals as follows: (1) academic medical centers, as defined by the Association of American Medical Colleges (with 3.3% of sample stroke patients); (2) other COTH institutions (8.8%); and (3) all other teaching hospitals, ie, those that have residency programs but are not COTH members (27.4%). Nonteaching hospitals constituted a fourth category (with 60.5% of sample patients).
Defining Costs and Outcomes
Costs were calculated from all claims associated with the hospitalization and the post–acute-care period (up to 90 days after admission). For the most part, these are costs and not Medicare payments. We converted all facility charge data to costs, using departmental cost-to-charge ratios calculated from each institution's Medicare Cost Report. These cost reports were obtained for acute-care hospitals, rehabilitation and other long-stay hospitals, and SNFs. No such cost-reporting data are available for other Medicare services (ie, physicians and suppliers, outpatient departments, and home health care); we simply used Medicare's allowed payments for these cases. We then adjusted all costs for geographic price differences using the Prospective Payment System wage index.
Outcomes included mortality and discharge destination. Mortality rates were calculated for the 90-day period after hospital admission. The patient's destination at discharge (eg, specialized rehabilitation facility, SNF, home) was determined from claims data.
The primary statistical analysis involved a comparison of costs and outcomes across the five specialties. We compared continuous variables using ANOVA and compared categorical variables using χ2 testing. When the overall comparison indicated differences among the five specialties, we followed up with multiple pairwise tests (eg, neurologist versus internist, neurologist versus FP). When multiple comparisons are made, the number of statistically significant results increases with the number of tests, even in the absence of actual differences. For this reason, some statisticians recommend setting a more stringent criterion for statistical significance for multiple pairwise comparisons (eg, P<.01). To accommodate this, we report all pairwise tests at both the P=.05 and P=.01 levels. For categorical variables with more than two categories (eg, discharge destination), we used a single χ2 test to determine whether there were any differences in the overall pattern of categories (taken as a whole) across physician specialties.
Multivariate regression analysis was used to adjust for all of the patient characteristics (available from claims data) that might affect cost and outcome differences. We estimated total episode costs in logged form using ordinary least squares, and we estimated mortality rates using logistic regression. Patient covariates included age (logged in the cost regression and included in both linear and squared form in the mortality regression), sex, race, comorbidity, whether the patient was transferred from another acute-care hospital, whether the patient originally became eligible for Medicare because of disability, prior history of strokes or TIAs, and whether the patient died during the episode (the latter was included in the cost regression only). Because neurologists practice disproportionately in teaching hospitals, which historically have higher costs regardless of physician specialty, we included categorical variables for the three types of teaching hospitals. Categorical variables were also included for physician specialty; “other specialists” constituted the reference group.
Approximately 1 of every 9 Medicare patients (11.3%) admitted for stroke had a neurologist as an attending physician (Table 1⇓). An additional one fourth of patients (25.3%) were treated by both a neurologist and a primary care physician. Almost one half of stroke patients were treated by a primary care physician only, either internist (28.6%) or FP (18.6%). An attending physician could not be identified for a small number of stroke patients (4.1%), a proportion comparable to that found in studies of patients with many different diagnoses.4 Prior research has shown that these patients are somewhat more likely to be treated in teaching hospitals.14
As expected, the probability of being cared for by a neurologist was considerably higher in teaching hospitals, especially in academic medical centers. Still, approximately one half of stroke patients treated in academic medical centers did not have a neurologist involved in their care as an attending physician.
Costs and Outcomes
Table 2⇓ compares costs and outcomes across the five types of attending physician. Neurologists treating stroke patients were significantly more expensive than other physicians: 34% more expensive than FPs and approximately 22% more than internists and other specialists. Furthermore, costs were higher for all parts of the episode: the hospitalization itself, inpatient physician services, and after the acute-care period. However, stroke patients treated by a combination of neurologists and primary care physicians were more expensive still, averaging almost $1000, or 5%, more than patients treated by a neurologist alone.
Stroke patients treated by neurologists achieved dramatically better outcomes than those treated by other specialists. Ninety-day mortality rates for patients cared for by neurologists were significantly lower than those for other attending physicians: 31% lower than those for internists, for example, and 36% lower than those for FPs. Although neurologist/primary care combinations averaged higher stroke mortality rates than neurologists alone (19.4% versus 16.1%), their mortality rates remained significantly lower than those of internists or FPs (based on significance tests not shown in this report). Stroke patients treated by neurologists also are more apt to be discharged to inpatient rehabilitation facilities or to be sent home compared with other stroke patients, suggesting that functional outcomes also may have been better. By contrast, stroke patients treated by other physicians were significantly more likely to be discharged to SNFs or nursing homes.
Utilization of In-Hospital Services
What were neurologists doing differently with stroke patients during the hospitalization? Table 3⇓ compares hospital length of stay and the use of selected cerebrovascular tests and other services. Stroke patients treated by neurologists do stay in the hospital longer, which may be one factor in their higher costs. (Their greater use of specialized rehabilitation facilities is probably another cost factor.) Neurologists also were significantly more likely to order diagnostic cerebrovascular tests: more head CT scans, more MRI scans of the brain, more cerebral angiography, and more noninvasive testing of the carotid arteries. The differences in use were particularly marked for MRI scans. Treatment by a combination of neurologists and primary care physicians generally resembled that of neurologists alone.
To the extent that neurologists may be more likely to practice in teaching hospitals that have more of this cerebrovascular technology, this might explain their greater use by neurologists. This cannot be the explanation in academic medical centers and other COTH institutions, however; virtually all of these institutions have these technologies, yet they still were being used more frequently by neurologists than by other physicians (Table 4⇓).
We were particularly interested in the use of warfarin in the treatment of stroke patients (for prevention of recurrence). Unfortunately, the Medicare hospital claims contain no information on the types of drugs used, and prescription drugs are not covered outside the hospital. Patients on warfarin should be monitored with prothrombin time tests, however, and we used outpatient bills for these tests within the first 30 days of discharge as a proxy for warfarin use. Because patients discharged to another institution do not have itemized laboratory bills, our measure of warfarin use was calculated only for patients discharged home. This represented approximately one half of all stroke patients treated by either neurologists or nonneurologists. Patients with a neurologist involved in their care (either as the sole attending physician or in combination with a primary care physician) were significantly more likely to be on warfarin than other patients.
Patient Case Mix
Patient case mix could be an important factor in these cost and outcome differences if neurologists tend to treat less seriously ill stroke patients or those patients with a greater chance of recovery. Neurologists definitely treated a younger mix of stroke patients (Table 5⇓). Although on average their stroke patients were only approximately 2 years younger, the percentage who were very old was markedly different. Neurologists were significantly less likely than other physicians to treat stroke patients aged 80 years and older. Neurologists also were more likely to treat male stroke patients, in large part because of their relatively younger age mix. (The proportion of Medicare patients who are male declines markedly with age.) There were no systematic differences in patient race between neurologists and other specialists.
We sought to identify comorbid illness among stroke patients that might suggest poor prognoses. Table 6⇓ displays the frequency distribution of Charlson Index scores for the eight groups of patients. The Charlson Clinical Comorbidity Index is the sum of selected chronic conditions shown to be associated with poor outcomes.15 There were no differences in mean index scores for stroke patients treated by neurologists compared with internists or other specialists. Stroke patients treated by FPs were less seriously ill as measured by the Charlson Index, however.
The Charlson Index was originally developed from a cohort of medical (unspecified) patients and then tested in a cohort of breast cancer patients.16 It was later adapted for use with claims data and validated on a sample of patients undergoing lumbar spine surgery.15 To better capture comorbidity specifically associated with stroke, we identified six comorbid conditions expected to increase the risk of poor outcomes for patients with cerebrovascular disease: hypertension, congestive heart failure, diabetes mellitus, acute myocardial infarction, chronic obstructive pulmonary disease, and valvular heart disease. The percentage of patients with each of these six conditions is also shown in Table 6⇑. Although there were some differences, they were not consistent between neurologists and other physicians. Stroke patients treated by neurologists were generally more likely to have secondary diagnoses of hypertension and valvular heart disease, for example, but less likely to have congestive heart failure or diabetes, compared with patients treated by other physicians. All differences are relatively small, however. It should be noted, furthermore, that all but two of these conditions (hypertension and valvular heart disease) are also included in the Charlson Index.
Stroke patients with prior cerebrovascular admissions may be more difficult to treat if they have residual neurological deficits. If so, this can explain only part of the observed cost and outcome differences between neurologists and nonneurologists. While stroke patients treated by neurologists did have significantly fewer prior admissions for cerebral infarction or TIA, the absolute differences were quite small.
Stroke patients may be selectively triaged to neurologists on the basis of their perceived severity. If so, then a greater number of cases treated by neurologists should be transfers from another acute-care hospital. Prior research has shown that the majority of transfers are from smaller, community hospitals to larger, more technology-intensive hospitals.17 While neurologists were significantly more likely than internists to receive stroke cases as transfers, the absolute number of transfers was quite small: 2.2% versus 1.5%. Furthermore, there were no differences between neurologists and FPs or between neurologists and other specialists.
Adjusting for Patient Case Mix
Table 7⇓ displays predicted costs and outcomes for attending physician specialty, adjusted for all of the patient and hospital characteristics, based on multivariate regression analysis. (We also had tested for interaction effects between physician specialty and hospital type, and between physician specialty and patient age, in both the cost and mortality regressions. The impact of physician specialty did not vary by type of hospital or patient age in either regression.) Neurologists remain significantly more expensive in their treatment of stroke patients, although the size of the cost differential has narrowed, averaging 18% more than internists and 31% more than FPs. The cost differential between patients treated by neurologists working alone and those working in combination with primary care physicians actually widens, however; combination cases cost 12% more to treat than those with a neurologist alone as the attending physician.
Neurologists still achieve significantly better outcomes than all other specialty groups. Stroke patients treated by a neurologist are 22% less likely to die within 90 days than an otherwise similar patient treated by an internist. Although patients treated by neurologist/primary care combinations have higher mortality rates than those treated by neurologists alone, their mortality rates still are significantly lower than those of FPs or other specialists. The mortality rate difference between internists and combinations is not quite significant at conventional levels (P<.10).
Role of Triaging
Triaging of patients to neurologists, based on clinical characteristics not observable with claims data, may explain their lower mortality rates. If so, then we would expect such triaging to occur more frequently in areas with relatively fewer neurologists. To investigate this, we classified all patients into one of four categories on the basis of the state's neurologist-population ratio: 4.0 or more, 3.0 to 3.9, 2.0 to 2.9, and less than 2.0 per 100 000 population. The percentage of stroke patients treated by neurologists ranged from a high of 13.2% in the most neurologist-rich states to a low of 7.8% in the most neurologist-poor states.
Table 8⇓ compares mortality rates across the five types of attending physician within each of the four neurologist-population categories. Stroke patients treated by neurologists achieved better outcomes in all four categories, but the magnitude of the difference is greater in the neurologist-poor states. The neurologist-internist differential is 6.4 percentage points in states with four or more neurologists per 100 000 population, for example, but 10.1 percentage points in states with fewer than two neurologists. While suggestive of triaging, a test of this relationship (in which interaction terms for physician specialty and state neurologist-population category were used) was not significant at the .05 level. We may simply have lacked the power to detect differences by state category. It also is possible that the most seriously ill stroke patients are less likely to be treated by neurologists in neurologist-poor states because they die before they can be referred to a neurologist. If so, then all of these early deaths would be attributed to primary care physicians. We replicated Table 8⇓, excluding all deaths within the first 2 days of admission. Although the means varied slightly, the relative differences between neurologists and other physicians remained unchanged in each of the four state groups (data not shown).
Our results suggest that stroke patients treated by neurologists achieve better outcomes, albeit at higher costs. The claims data used for this study included only limited information on stroke severity and comorbidity; it is possible that unmeasured clinical characteristics might explain the observed differences for neurologists and other physicians. Similarly, our outcome measures were limited to mortality and to destination on discharge from the hospital, with no direct measure of functional status. Nevertheless, our results are suggestive, especially when coupled with similar findings from randomized controlled trials in Finland and Scandinavia.8 9 10
How can we explain better outcomes for stroke patients treated by neurologists? There are several dimensions of stroke management that may be relevant: diagnostic workup, treatment, prevention of complications, and rehabilitation. While we cannot observe the process of care from claims data, we do know that patients treated by neurologists were significantly more likely to receive diagnostic cerebrovascular studies of all kinds, especially MRI scans of the brain. The increased use of these tests may have provided important clinical information aimed at determining the pathophysiological mechanism of stroke, which then affected the course of treatment. We did observe, for example, that patients treated by neurologists were more apt to be on warfarin after discharge. In addition, we know that patients treated by neurologists were more likely to be discharged to specialized rehabilitation facilities, suggesting that they may have been more aware of the value of early rehabilitation. The neurologist as specialist may be a marker for better identification of mechanism of stroke, focused nursing care, and a team approach to the care of patients with stroke.
More research on stroke treatment differences by physician specialty clearly seems warranted. Are the lower mortality rates achieved by neurologists in fact attributable to the care provided by these specialists, or is there systematic triaging to neurologists based on clinical characteristics we could not observe in our study? A definitive answer to this question can be answered only through a randomized clinical trial. While such a trial may be impossible in today's fiscal climate, Kassirer18 has warned that, as the pressure increases to substitute generalists for specialists, it is important to determine the optimal use of specialists for specific medical conditions. The greater use of warfarin by neurologists suggests that these specialists may have been more knowledgeable about its benefits for prevention of stroke recurrence, for example, than primary care physicians. This finding is consistent with the greater knowledge by cardiologists of the appropriate drug therapies for treating another life-threatening illness: acute myocardial infarction.5 If so, then broader dissemination of stroke treatment guidelines among primary care physicians may be warranted.
Selected Abbreviations and Acronyms
|COTH||=||Council of Teaching Hospitals|
|ICD-9||=||International Classification of Diseases, 9th Revision|
|PORT||=||Patient Outcomes Research Team|
|SNFs||=||skilled nursing facilities|
|TIA||=||transient ischemic attack|
This study was supported by contract No. 282-91-0028 from the Agency for Health Care Policy and Research (Stroke Prevention PORT).
Reprint requests to David B. Matchar, MD, Center for Health Policy Research and Evaluation, Duke University, Erwin Square, Suite 230, 2200 W Main St, Durham, NC 27705.
- Received May 22, 1996.
- Revision received July 18, 1996.
- Accepted July 18, 1996.
- Copyright © 1996 by American Heart Association
Physician Payment Review Commission. Annual Report to Congress 1993. Washington, DC: Physician Payment Review Commission; 1993.
Roback G, Randolph L, Seidman B. Physician Characteristics and Distribution in the U.S. Chicago, Ill: American Medical Association; 1993.
National Center for Health Statistics. Health United States 1992. Hyattsville, Md: Public Health Service; 1993.
Indredavik B, Bakke F, Solberg R, Rokseth R, Haaheim LL, Holme I. Benefits of a stroke unit: a randomized controlled trial. Stroke. 1991;22:1026-1031.
Strand T, Asplund K, Eriksson S, Hagg E, Lithner S, Wester PO. Stroke unit care: who benefits? Comparisons with general medical care in relation to prognostic indicators on admission. Stroke. 1986;17:377-381.
Kaste M, Palomaki H, Sarna S. Where and how should elderly stroke patients be treated? A randomized trial. Stroke. 1995;26:249-253.
American Academy of Neurology. Neurologists 1991-1992. Minneapolis, Minn: American Academy of Neurology; 1993.
Leibson CI, Naessens JM, Brown RD, Whisnant JP. Accuracy of hospital discharge abstracts for identifying stroke. Stroke. 1994;25:2348-2355.
Buczko W. Inpatient transfer episodes among aged medicare beneficiaries. Health Care Financing Rev.. 1993;15:71-87.