Background

Ischemic heart disease is the leading cause of death worldwide, causing 6.26 million deaths per year[1]. Acute myocardial infarction (AMI) is a major cause of death in the United States, accounting for 203,551 deaths in 1998[2]. A growing number of AMI patients are transferred from one hospital to another during their hospital course[3, 4].

While there is nothing intrinsically beneficial about moving a patient from one hospital to another, transfer may provide the opportunity for higher level of care and more advanced treatment. Several observational studies on general medical and surgical patients reported that transferred patients, regardless of their diagnosis, were sicker, had more co-morbid conditions, used more resources, required longer hospital stays, and had higher mortality [5–7]. Another study found that transferred patients had less severe illness and lower mortality[8]. Factors that might confound these previous findings include changing economic motivation for transfer, greater variation in availability of advanced technology, and widespread attempts to improve quality of care [3, 9].

Early studies on myocardial infarction called for the transfer of "high risk patients"[10]. The conventional wisdom was to transfer the sickest cardiac patients or patients who had failed less invasive therapy to the tertiary care hospital for specialized care [11–14]. However, several more recent studies found that transferred acute MI patients were younger and had fewer comorbid conditions[4, 15]. Rural MI patients are more likely to be transferred, however, rural patients have also been reported to receive lower quality of care[16]. Many studies on acute MI have deleted some or all transferred patients from their analysis [17–19].

The impact of interhospital transfer on processes and outcomes of acute MI has largely gone unstudied. Because the number of AMI patients undergoing interhospital transfer is rising we examined the impact of interhospital transfer on mortality. We used data from the Cooperative Cardiovascular Project (CCP), a large and representative sample with detailed clinical and quality of care information on patients hospitalized with AMI.

Methods

Cardiovascular Cooperative Project

The CCP was a national quality improvement project sponsored by the Centers for Medicare and Medicaid Services (CMS), formerly the Health Care Financing Administration for Medicare patients hospitalized with AMI[20, 21]. Patients were initially identified from Medicare claims data using the principal diagnosis code of 410 from International Classification of Diseases, Ninth Revision, Clinical Modification[22]. The CCP performed structured medical record review for 234,769 Medicare fee for service patients randomly sampled from 6,684 hospitals in all 50 states who were hospitalized for AMI between February 1994 and July 1995. As a quality check, an independent abstraction for a randomly selected 5% of the charts was done to assess reliability and validity for key variables. The methods of the CCP are fully described elsewhere[20, 21].

Patients

Patients were excluded from our analyses for: 1) lack of clinically confirmed AMI according to criteria established by Ellerbeck[21] (n = 31,194); 2) admission to hospital with unclear teaching status, technology index, or rural/urban status (n = 262); 3) age less than 65 years (n = 15,072); and 5) death on day of admission for patients who were not transferred (n = 3,946). Patients who died on the day of admission were excluded because they had less opportunity for transfer. Application of these exclusions left 184,295 patients from 4,765 hospitals.

Patients who underwent transfer were our primary study group of interest. We define "transfer" as occurring when a patient is admitted to one acute care hospital and discharged from a different hospital during an episode of care for an AMI. The CCP records for patients transferred into an index hospital contain the source of admission but no detailed information on pre-admission clinical course. For patients transferred from an index hospital, the records contain discharge destination but not detailed information on post-discharge course. However, vital status was available for all patients from Medicare administrative data, in particular at 30 days after hospitalization.

Hospitals

Teaching hospitals were defined as those with an intern/bed ratio greater than zero[8], as assessed by merging the CCP and CMS administrative data sets. Hospital location was defined as rural or urban by metropolitan statistical area[23].

For each hospital, we derived a technology index (TI) as described below:

  1. 1.

    No Angiographic, PTCA or CABG capacity

  2. 2.

    Angiographic capacity only

  3. 3.

    Angiographic and PTCA capacity only

  4. 4.

    Angiographic, PTCA, and CABG capacity

Process and outcome measures

We focused on four quality measures that were developed as part of the CCP: in-hospital administration of aspirin, beta-blockers, ACE- inhibitors and acute reperfusion to eligible candidates. We considered beta-blockers indicated for those patients who both lacked absolute contraindications and met inclusion criteria[20]. We considered aspirin and thrombolytic therapy indicated in patients who lacked both absolute and relative contraindications and met inclusion criteria. Definitions of contraindications and therapy groups are described further in Allison et al[24]. Because of significant interactions between receipt of acute reperfusion and aspirin, we created 5 mutually exclusive groups of patients: therapy category 0, those who received no therapy (reference group); therapy category 1, those who received no aspirin and no reperfusion but did receive angiotensin-converting enzyme (ACE) inhibitors and/or b-blockers; therapy category 2, those who received no reperfusion but did receive aspirin and/or ACE inhibitors and/or b-blockers; therapy category 3, those who received no aspirin but did receive reperfusion and/or ACE inhibitors and/or b-blockers; and therapy category 4, those who received aspirin and reperfusion and/or ACE inhibitors and/or b-blockers.

Patients transferred in from another acute care hospital were not considered eligible for reperfusion at the receiving hospital. Mortality at 30 days after admission was ascertained from HCFA administrative data. Patient co-morbidity and severity of illness were assessed by the Adapted Charlson Index[25] and by the APACHE II scale[26], respectively.

Statistical analyses

Two separate analyses were conducted. First, we compared transferred and non-transferred patients. Because interhospital transfer is more common in rural hospitals we performed a second analysis comparing patients cared for in rural and urban hospitals. Differences between groups (transferred v non-transferred and rural v urban) were tested using the chi-square or Kruskal-Wallis test[27].

We first examined unadjusted 30-day mortality for each comparison group; transfer v non-transfer, and rural v urban. Predicted mortality was calculated using patient demographics, and other clinical predictors of mortality based on the work of Krumholz (age, gender, race, serum white blood count [WBC] on admission, serum creatinine, presence of heart failure on admission, cardiac arrest, and location of MI)[28].

Because transfer status and location of hospital was not randomly assigned in this patient population, there was substantial potential for confounding and selection bias. We chose to account for this by developing a propensity score for transfer and rural status. Propensity score analysis is a post-hoc statistical method that estimates treatment effect when subjects were not randomly assigned to treatment group. Propensity score analysis attempts to simultaneously control for all known patient factors that might be related to the outcome of interest. Joffe and Rosenbaum have described the rationale and methods underlying the use of propensity score analysis[29].

We constructed nonparsimonious logistic regression models in which inter-hospital transfer was a dependent variable and the variables in Table 1 were independent variables. These models made it possible to calculate a propensity score, indicating the likelihood that any individual patient would undergo interhospital transfer given all other known variables except 30-day mortality. The C-statistic of the logistic regression model used to generate the propensity score for transfer was 0.68 indicating a moderate ability to differentiate between transferred and non-transferred patients. The C-statistic represents the discriminative power of the logistic regression model. We performed a similar analysis for patients cared for in a rural hospital. The C-statistic of the logistic regression model used to generate the propensity score for rural hospital status was 0.57 indicating a fair ability to differentiate between rural and urban patients.

Table 1 Patient characteristics for transferred v non-transferred patients and rural v urban patients*
Full size table

We used the propensity score to randomly match transferred patients to non-transferred patients and rural patients to urban patients. Specifically, pairs of propensity scores were randomly matched using a greedy matching technique[30]. This technique randomly matched 44,175 transferred and non-transferred patients and 32,131 rural and urban patients.

For each comparison we then used a multivariable logistic regression model to adjust mortality analyses for patient demographics, severity of illness based on work by Krumholz[28], treatment according to a schema developed by Allison[24], hospital technology index, teaching status, and hospital size. We included in each analysis any additional covariant for which there had been a significant difference among our propensity-matched cohorts. All analyses were conducted using SAS version 9.00 (SAS Institute Inc., Cary, NC).

Results

Hospital and patient characteristics

Our study sample consisted of 184,295 patients from 4,765 hospitals. Overall, 51,530 AMI patients (28.0%) were transferred at some point during their hospital stay; 32,080 (17.4%) were transferred out of and 19,450 (10.6%) transferred into a CCP hospital. 35,824 (19.4%) patients were cared for in a rural hospital. Rural hospitals transferred 33.8% while urban hospitals only transferred 26.6%. Hospitals with lower technology index transferred a higher proportion of their patients (34% for low technology hospitals v 23% or high technology hospitals).

Baseline patient characteristics according to transfer status and hospital location are summarized in Table 1. Because the CCP is such a large database there are statistically significant differences that do not necessarily represent clinically significant differences. It is essential to consider both clinical and statistical significance when reviewing these tables. Table 2 describes treatment and crude mortality rates.

Table 2 Treatment and crude mortality among transferred v non-transferred patients and rural v urban patients
Full size table

Transfer and mortality

Transferred patients were younger and had much lower predicted mortality than non-transferred patients. Transferred patients were more likely to be male, less likely to be African American, and less likely to have diabetes and a history of heart failure. Transferred patients were less likely to have heart failure or an abnormal heart rhythm on admission. Transferred patients had higher rates of use for of aspirin, beta-blockers, and thrombolytics (Table 2).

30,586 (16.6%) patients died within 30 days. Unadjusted 30-day mortality was lower among transferred patients (10.7% v 18.9%, p < .001). Based on systematically collected data for baseline demographics and medical risk factors a logistic regression model was used to generate a propensity score for transfer. The baseline demographic and medical risks comparing propensity matched transferred and non-transferred patients are shown in Table 3. As opposed to the entire population of CCP patients, these propensity-matched patients were well matched; the only clinically significant differences were that transferred patients had a slightly higher predicted mortality (15% v 14%, p = .007) and a slightly lower rate of diabetes (29.5% v 30.3%, p = .009). Transferred patients had a lower unadjusted 30-day mortality (10.7% v 12.5%. p < .001). This mortality advantage persisted after adjustment for patient demographics, therapy, and hospital characteristics. (O.R. for 30-day mortality = 0.80, 95% C.I. 0.76–0.84) (Table 4).

Table 3 Patient characteristics according to transfer status and hospital location in propensity matched groups
Full size table
Table 4 Odds ratios for 30-day mortality in propensity score matched patients across model groups
Full size table

Rural hospital and mortality

35,824 (19.4%) patients were cared for in a rural hospital. Rural patients were slightly older than urban patients, but had similar predicted mortality (Table 1). Rural patients were more likely to be female, less likely to be African American, had a higher rate of history of heart failure but had similar rates of diabetes. Rural patients were slightly less likely to have heart failure or an abnormal heart rhythm on admission. Rural patients had lower rates of use for aspirin, beta-blockers, and ACE-Inhibitors, but a higher rate of use for thrombolytic therapy (Table 2).

The baseline demographic and medical risks among propensity matched rural and urban patients are shown in Table 3. These propensity-matched patients were well matched. Rural patients had a higher unadjusted 30-day mortality (17.5% v 16.3%. p < .001). However, after adjustment for patient demographics, therapy, and hospital characteristics this mortality advantage disappeared. (O.R for 30-day mortality = 1.05, 95% C.I. 0.99–1.11) (Table 4).

Discussion

We found that over one quarter of AMI patients were transferred during their hospital course. Transferred patients were significantly younger and more often white, male, with fewer co-morbid conditions and less severe disease. Transferred patients were also more likely to receive appropriate therapy. Transferred patients had lower unadjusted 30-day mortality than non-transferred patients. After a rigorous propensity analysis of nearly 90,000 propensity -matched patients, this mortality benefit persisted. Physicians decide to transfer patients for many explicit and implicit reasons. The transferring physicians in the CCP may have understood which patients would most benefit from transfer. Testing these hypotheses will require prospective data collected with more clinical detail than we have available in the current data.

We found that patients cared for in rural hospitals had slightly lower rates of treatment with 3 quality care measures (aspirin, beta-blockers, and ACE-Inhibitors), while they had a higher rate of treatment with thrombolytic therapy. This may be due to decreased availability of angioplasty in rural hospitals. After analysis of over 64,000 propensity-matched patients and adjustment for patient differences, treatment differences and hospital characteristics, patients cared for in a rural hospital had similar mortality as patients cared for in an urban hospital. The high rate of transfer among younger, healthier patients may partially account for the lower quality of care and worse outcomes ascribed to rural hospitals previously reported by others.

To our knowledge, this is the first national study of the characteristics and mortality of Medicare patients with AMI according to transfer status. Our results are consistent with those of Mehta and colleagues, who used CCP data only from the state of Michigan to examine the implications of patient transfer[15]. Mehta found that patients who were transferred from hospitals with lower technological capability to hospitals with higher technological capability tended to be younger, more likely to be white and male, and had lower predicted mortality.

The bulk of the literature on transfer of patients with AMI focuses exclusively on patients transferred for specific procedures [31–34]. For example, Straumann et al. evaluated prospectively the baseline characteristics and outcomes of AMI patients transferred to a tertiary referral center for primary PTCA and compared these patients with patients directly admitted to the same referral center[35]. They found that the patients who were transferred-in tended to be younger, more critically ill, more likely to be in cardiogenic shock or require resuscitation, but had similar mortality. Liem et al. compared transferred to non-transferred PTCA patients to evaluate treatment delay, infarct size and mortality[36]. They found that despite an average 43-minute treatment delay for transfers and larger infarct size, transferred and non-transferred patients had similar 6-month clinical outcomes. Andersen et al. recently reported that patients transferred for primary PTCA had better outcomes than non-transferred patients receiving thrombolytic therapy[37]. However, the benefit was solely in terms of decreased re-infarction and there was no statistically significant benefit to transfer in terms of mortality or stroke.

Our finding that there are major differences between transferred and non-transferred patients has particular relevance for the understanding of quality of care in rural hospitals. Previous studies have frequently deleted transferred patients from analysis. Because transferred patients tend to be younger, healthier, male, and have lower predicted mortality, comparisons between hospitals are subject to a significant bias against hospitals that transfer a higher proportion of AMI patients. Thiemann and Casale in separate reports found that rural and smaller hospitals had worse outcomes than urban and larger hospitals[16, 17]. However, their studies deleted transferred patients from their analysis. After accounting for the numerous and large differences between transferred and non-transferred patients, we found that patients cared for in rural hospitals had similar outcomes to patients cared for in urban hospitals.

Although the reason for transfer is not documented in the CCP dataset, we can make some inferences based on the characteristics of hospitals transferring and accepting patients, and by the treatments administered to each group of patients. From our data it is clear that smaller, rural hospitals with less technological capacity are more likely to transfer patients to another institution. Larger, urban hospitals with the ability to perform cardiac catheterization, PTCA, and bypass surgery are less likely to transfer.

In a rural hospital without advanced cardiac services transfer may be viewed as a treatment option, just like the use of aspirin, beta-blockers, and thrombolytics. While there are evidence-based guidelines for the medical treatments of acute MI, there are no guidelines aiding the decision of whether to transfer a patient suffering an acute MI. Certainly, a patient who requires cardiac surgery or urgent angiography will benefit from transfer. Identification of other patient groups likely to benefit from transfer will provide guidance to the clinician faced with the decision whether or not to transfer a patient.

For health services research the issue surrounding the analysis of transferred patients is complex. There is disagreement about where to assign responsibility for outcomes. Because transfer is so common and may actually represent a treatment option, rather than an outcome, the "assignment of responsibility" may not be the most important question. The important question for hospitals without interventional cardiac services may be how to identify the patient who is most likely to benefit from transfer. Transfer rates have increased dramatically in the past decade making it even more important to understand the risks and benefits associated with transfer[4].

The major limitation of this study is that transfer and rural hospitalization were not randomly assigned. The use of observational studies to assess treatment effects and outcomes is controversial. Additionally, chart review has its own unique limitations[38]. Several recent publications point out that properly performed observational studies are unlikely to lead to misleading or inappropriate conclusions[38, 39]. We performed propensity analysis that provided a robust adjustment for selection bias and confounding. However, propensity analysis can only adjust for measured variables. For example, the CCP does not collect data on socioeconomic factors that may be related to use of more invasive treatment[40] and may predispose patients to transfer as well as to improved survival. Due to missing data we were unable to match a small portion of the transferred and rural patients. However, our match rate of 86–90% is within the range of previously reported propensity analyses[41, 42].

Conclusion

We found that Medicare patients that underwent interhospital transfer during care of their acute MI had generally higher quality of care and lower mortality than non-transferred patients. However transferred patients were very different than non-transferred patients. These differences may partially account for the difference in mortality between rural and urban hospital previously reported. We found no difference in 30-day mortality between patients cared for in a rural or urban hospital. Deleting transferred patients from analysis may introduce significant bias. For patients living in a rural community it is reasonable for them to present to their local hospital, and the decision whether to transfer or not becomes a clinically important element in their care. Additional work is needed to better define characteristics of the patient and the health care system that might identify those most likely to benefit from transfer and methods to expedite transfer for those patients.