Kansagra SM, Curtis LH, Schulman KA. Regionalization of Percutaneous Transluminal Coronary Angioplasty and Implications for Patient Travel Distance. JAMA. 2004;292(14):1717-1723. doi:10.1001/jama.292.14.1717
Author Affiliations: Center for Clinical and
Genetic Economics, Duke Clinical Research Institute, and Duke University School
of Medicine, Duke University Medical Center (Drs Curtis and Schulman, and
Ms Kansagra), and Health Sector Management Program, The Fuqua School of Business,
Duke University (Dr Schulman and Ms Kansagra), Durham, NC.
Context Minimum procedure volume thresholds have been proposed to improve outcomes
among patients undergoing percutaneous transluminal coronary angioplasty (PTCA).
How regionalization policies would affect patient travel distances is not
Objective To examine the effect of regionalization of PTCA on patient travel distances.
Design, Setting, and Participants A retrospective cohort study of discharge records, which were examined
to determine hospital and operator procedure volumes, of 97 401 patients
undergoing PTCA in New York, New Jersey, and Florida in 2001. Travel distances
were measured at baseline and under 2 regionalization scenarios in which hospital-operator
pairs not meeting minimum volume standards stopped providing services.
Main Outcome Measures Observed and expected patient travel distances, and risk-adjusted mortality.
Results With a minimum volume standard of 175 per operator and 400 per hospital
(class 1), 25% of patients had a shorter travel distance, 10% had a longer
travel distance, and 65% experienced no change. Most patients with longer
travel distances under this standard would travel no more than 25 miles farther,
and most patients with shorter travel distances would save no more than 10
miles. With a minimum volume standard of 75 per operator and 400 per hospital
(class 2), 11% of patients had a shorter travel distance, 2% had a longer
travel distance, and 87% experienced no change. Under both standards, less
than 1% of patients would travel more than 50 miles farther than their observed
travel distance. Risk-adjusted mortality was higher for lower-volume hospital-operator
pairs (1.2% for class 3 vs 0.9% for class 2 and 0.8% for class 1; P<.001 for both comparisons).
Conclusion Regionalization of PTCA would not increase travel distance for most
patients; however, potential costs of regionalization not related to travel
must be examined before such policies can be recommended.
The relationship between procedure volume and patient outcomes is well-established
for percutaneous transluminal coronary angioplasty (PTCA).1- 7 At
operator and hospital levels, higher procedure volume is associated with lower
rates of inpatient mortality, emergency coronary artery bypass graft surgery,
and complications.1- 3,5- 8 Minimum
procedure volume thresholds have been proposed to improve outcomes among patients
undergoing PTCA. Jollis et al1 estimated significant
improvements among patients whose operators performed at least 150 procedures
annually and at hospitals with at least 400 procedures annually. The American
College of Cardiology recommends that hospitals perform at least 400 procedures
annually and that operators perform at least 75 procedures annually.9 The Leapfrog Group, a coalition of health care benefits
organizations, encourages purchasers of health care to direct patients to
hospitals that perform more than 400 procedures annually.10
Although regionalization of PTCA may reduce morbidity and mortality
by diverting patients away from low-volume hospitals and operators, it may
also affect access to care as low-volume centers in remote or rural areas
stop providing services. Birkmeyer et al11 examined
this issue with respect to pancreatectomy and esophagectomy and found that
low minimum volume standards would require approximately 15% of patients to
change hospitals. Of those patients, 75% would have traveled no more than
30 additional minutes. Pancreatectomy and esophagectomy are relatively
uncommon procedures. It is unclear how regionalization would affect patient travel
burden for more common procedures that often are performed on an urgent or
We sought to determine the impact of 2 hypothetical minimum volume standards
for PTCA on patient travel distances. We also conducted an exploratory analysis
of patient characteristics associated with observed travel distances and,
for patients who did not seek care from the closest hospital, characteristics
associated with travel beyond the closest hospital. As a secondary outcome,
we calculated the risk-adjusted mortality associated with hospital-operator
pairs in different volume categories.
Data were from the State Inpatient Databases of the Healthcare Cost
and Utilization Project, which comprised administrative databases containing
inpatient discharge abstracts from nonfederal US hospitals in participating
states. We used 2001 data from 3 states (New York, New Jersey, and Florida),
which included surgeon identifiers, hospital identifiers, and 5-digit patient
and hospital ZIP codes necessary for the analysis. We evaluated patients in
New York and New Jersey as a single group because they were close enough to
seek care at a hospital in either state. We did not evaluate patients who
may have sought care in other neighboring states because data were not available
for those states.
The study population included all patients with a documented PTCA as
indicated by International Classification of Diseases, Ninth
Revision (ICD-9) codes 36.01, 36.02, and 36.05.
We calculated operator procedure volumes by summing procedures over unique
physician identifiers, and hospital procedure volumes by summing procedures
over unique American Hospital Association identification numbers. Hospitals
and operators with less than 5 procedures per year were excluded because of
possible errors in coding.2 Each procedure
was counted, even when PTCA was performed multiple times in the same patient
during the same hospitalization, which occurred in 3% of all discharges.1 For operators practicing at more than 1 hospital,
volume reflected all procedures performed in that state.
In addition to calculating observed travel distances, we calculated
expected travel distances for each patient under 2 hypothetical minimum volume
standards based on existing proposals1,3,9,10:
minimum annual operator volume of 175 at a hospital with a minimum annual
volume of 400 (standard 1) and minimum annual operator volume of 75 at a hospital
with a minimum annual volume of 400 (standard 2). To meet a minimum volume
standard, the hospital-operator pair had to meet both the operator dimension
and the hospital dimension of the standard. Hospital-operator pairs were classified
according to the highest standard met. Thus, hospital-operator pairs that
met standard 1 were considered to be class 1 organizations, pairs that met
standard 2 but not standard 1 were considered to be class 2 organizations,
and pairs that did not meet standard 2 were considered to be class 3 organizations.
The primary outcome measure was the effect of the hypothetical minimum
volume standards on patient travel distance. We determined geographic centroids
for patients and hospitals using a ZIP code centroid file (ZIPList5 Geocode,
CD Light LLC, The Woodlands, Tex). Geographic centroids correspond with the
geographic center of each ZIP code area. We assumed that all patients living
in a ZIP code area lived at the geographic centroid and patient ZIP codes
were based on the home address. We calculated straight-line travel distances
using the latitude and longitude of each pair of patient and hospital-operator
centroids with a formula (available upon request from the authors) that estimated
distance within 0.01 miles (0.02 km). We excluded patients who lived more
than 193 miles (309 km) from their hospital-operator (99th percentile) because
of the strong likelihood that these patients were not close to their place
of residence when they sought care. For each minimum volume standard, we calculated
the expected travel distance between the patient ZIP code and the ZIP code
of the nearest hospital-operator meeting the criteria of that standard. Patient
travel burden was defined as the difference between the expected travel distance
(under each minimum volume standard) and the observed travel distance. For
transfer patients, travel distance was measured from the patient’s home
address to the transfer facility. In a sensitivity analysis, patients from
the study population who were transferred to the hospital from other facilities
were removed because observed travel distances for these patients may not
accurately reflect the patients’ choice of hospital-operator pair.
We compared patient characteristics across hospital-operator classes
using χ2 tests for categorical variables and 1-way analysis
of variance for continuous variables. Statistical significance was defined
as P<.05, and all tests were 2-sided. Multivariable
log-linear regression was used to conduct an exploratory analysis on the independent
effect of patient characteristics on observed travel distance and distance
traveled beyond the closest hospital-operator pair. Patient characteristics
incorporated in the model were identified on the basis of prior studies and
clinical expertise and included age, sex, race/ethnicity, urgency of the procedure,
state of residence, myocardial infarction (MI) status, comorbidity, insurance
type, and admission source.1,2 For
the race/ethnicity variable, the Healthcare Cost and Utilization Project (hospital
discharge abstract) codes were used, which included black, white, Hispanic,
Asian, and other. Race/ethnicity was assessed in this analysis because previous
research has identified race/ethnicity as an important factor in the use of
coronary revascularization procedures. For the comorbidity variable, the D’Hoore
translation of the Charlson comorbidity index was used, which was developed
for use with ischemic heart disease–related ICD-9 codes listed among secondary diagnoses in administrative databases.12,13 The admission source variable indicated
whether the admission was routine (physician referral) from the emergency
department, a transfer from another hospital, or a transfer from another health
care facility. We converted distances of 0 to 0.00001 miles for the regression
To place our study population in the context of previous work demonstrating
the relationship between procedure volume and mortality,1- 7 we
examined the risk-adjusted mortality associated with each hospital-operator
class as a secondary outcome. A multivariable logistic regression model was
constructed using generalized estimating equations to account for the nonindependence
of patients within operators and hospitals. We excluded patients who were
transferred to another hospital because outcome data at the transfer facility
were not recorded in the database. Independent variables included patient
age, sex, race/ethnicity, comorbidity, state of residence, MI status, and
urgency.1,2 We then used the model
to compute predicted mortality for each patient. We summed the predicted mortality
of each patient to determine the expected number of deaths for each hospital-operator
class. We then multiplied the observed-to-expected ratio for each class by
the mortality rate for the overall population to determine the risk-adjusted
mortality for each class.14 Statistical analyses
were performed using SAS version 8.0 (SAS Institute, Cary, NC).
In 2001, 120 hospitals and 1418 operators performed PTCA on the 97 401
patients who were eligible for the study. A total of 23 148 patients
(24%) received care from hospital-operator pairs considered to be class 3;
these hospital-operator pairs failed to meet the hospital minimum volume standard
of 400, the operator minimum procedure volume of 75, or both (Table 1).
Table 2 summarizes patient characteristics
by hospital-operator class. Class 3 had the greatest percentage of patients
admitted through the emergency department and, consistent with this finding,
had the greatest percentage of procedures considered emergent. In addition,
a greater percentage of patients receiving Medicare sought care from class
1 pairs compared with class 2 or class 3.
Median (interquartile range [IQR]) travel distance was 9.5 miles (4.7-18.2
miles [15.2 km; 7.5-29.2 km]). A total of 51 280 patients (53%) traveled
less than 10 miles (<16 km) for the PTCA procedure, 41 157 (42%) traveled
10 to 50 miles (16 to 80 km), and 4964 (5%) traveled more than 50 miles (>80
Table 3 summarizes expected travel
distances under the hypothetical minimum volume standards. Under standard
1, 57 996 patients (60%) who received care from class 2 or class 3 pairs
would have been required to move to a class 1 hospital-operator. Paradoxically,
the median travel distance for these patients would have been shorter than
the observed distance. This occurred because many patients were traveling
to a low-volume hospital-operator pair when a higher-volume hospital-operator
pair often was closer. Approximately 10% of patients would have traveled farther
than their observed travel distance and 25% would have traveled a shorter
distance. Most patients with longer travel distances under this standard would
have traveled no more than 25 miles farther than their observed travel distance,
and most patients with shorter travel distances would have saved no more than
10 miles. Of the remaining 65% who would have experienced no change in travel
distance, 38% (25% of all patients) would have switched to a high-volume operator
at the high-volume hospital at which they underwent the procedure.
Under standard 2, 23 148 patients (24%) who received care from
a class 3 pair would have been required to switch to class 1 or class 2 pair.
The median (IQR) travel distance for these patients was 6.8 (3.3-14.2) miles
(10.9 [5.3-22.7] km). Class 1 was the closest alternative for approximately
79% of these patients and class 2 was the closest alternative for 21% of the
patients. Only 2% of patients would have traveled farther under this standard
and 11% would have traveled a shorter distance. Of the patients traveling
farther, most would have traveled no more than 25 miles farther than their
observed travel distance. For patients traveling a shorter distance, most
would have saved no more than 10 miles. The majority of patients (87%) would
have incurred no additional travel. Of these, 12% (11% of all patients) would
have switched to a high-volume operator in the same high-volume hospital.
Under both minimum volume standards, less than 1% of patients would
have traveled more than 50 miles farther than their observed travel distance.
In addition, under both minimum volume standards, the changes in travel distance
would have increased for more patients had all patients initially sought care
from the closest hospital-operator pair (Table
Younger patients, white patients, men, and patients not receiving Medicaid
were more likely to travel farthest for PTCA. Patients who transferred from
another hospital, patients without an MI, and patients admitted for an urgent
PTCA also traveled farther. However, the regression model explained only 5%
of the variance in observed travel distances. In the analysis of factors associated
with travel beyond the closest hospital-operator, younger patients, white
patients, men, patients not receiving Medicaid, patients without an MI, and
patients transferred from another hospital were more likely to travel beyond
the closest hospital-operator. Again, the regression model explained only
5% of the variance in travel distance beyond the closest hospital-operator.
However, unlike the analysis of factors associated with observed travel distance,
admission for an emergency procedure was also associated with greater travel
distance beyond the closest hospital-operator, most likely because patients
may not have been at home and were taken to the facility nearest their location.
Transfer patients had a median (IQR) travel distance of 12.2 (7.5-24.9)
miles (19.6 [12.1-40.1] km). In sensitivity analysis, in which transfer patients
were removed, median (IQR) observed travel distance was 8.6 (4.1-16.5) miles
(13.8 [6.6-26.6] km). Approximately 57% of nontransfer patients traveled less
than 10 miles, whereas only 39% of transfer patients traveled less than 10
miles. If nontransfer patients had initially traveled to the closest hospital-operator,
73% of these patients would have traveled less than 10 miles. Although nontransfer
patients had shorter median travel distances than transfer patients, more
than half of nontransfer patients were still not receiving care from the closest
In the mortality calculation, we excluded 6710 patients who were discharged
to another hospital and, therefore, could not be followed up. For the remaining
90 691 patients, unadjusted mortality increased with decreasing procedure
volume. Patients receiving care from class 3 pairs had higher mortality (1.3%)
than did patients receiving care from class 2 (0.9%) or class 1 (0.8%). Risk-adjusted
mortality followed a similar trend. Patients receiving care from class 3 pairs
had higher risk-adjusted mortality (1.2%) than did patients receiving care
from class 2 (0.9%) and class 1 (0.8%) (P<.001
for both comparisons).
Our findings suggest that regionalization of PTCA would not increase
travel distances for most patients in New York, New Jersey, and Florida. More
than half of patients did not travel to the closest institution for PTCA;
many bypassed hospitals with higher procedure volume and lower risk-adjusted
mortality. Under our hypothetical minimum volume standards, therefore, median
expected travel distances were shorter than observed travel distances. Observed
travel distances would have been shorter than expected travel distances had
all patients sought care from the closest hospital-operator.
We also found that insurance type, race/ethnicity, sex, age, and other
patient characteristics were significantly associated with observed travel
distances and with distance traveled beyond the closest hospital-operator.
Our model explained only 5% of the variation in observed travel distance,
suggesting that unmeasured factors account for most of the variation in observed
travel distance. However, the association between insurance status and observed
travel distances creates a concern that some health care plans may restrict
choice of hospital-operator and that patients in these plans may travel farther
despite having critical medical conditions. Time to the emergency department
influences outcomes in situations requiring immediate intervention.15,16 Because PTCA is often an emergency
procedure, increases in travel distance for some patients may not be worth
improvements in mortality for other patients. In our study, if the 23 000
patients who visited a class 3 hospital-operator had instead presented to
a class 1 hospital-operator, the associated 0.4% reduction in mortality would
have saved an additional 92 lives. However, this finding is valid only if
we assume that the greater mortality due to longer transit times for some
patients balances out the lower mortality due to shorter transit times for
other patients. Policies of regionalization must, therefore, take into account
factors, such as payer requirements that may affect travel distance, as well
as the effect of transit time on the outcome of the procedure itself.
To our knowledge, only 3 articles address the impact of regionalization
on travel distances. Grumbach et al17 conducted
a similar study of travel distances for patients undergoing coronary artery
bypass graft surgery. After hospitals performing less than 100 coronary artery
bypass graft procedures annually stopped providing services in a hypothetical
scenario, travel distances increased for less than 5% of patients. Birkmeyer
et al11 studied pancreatectomy and esophagectomy
and found that low minimum volume standards would have small effects on travel
time. Chang and Klitzner18 studied the effect
of regionalization on pediatric cardiac surgery and found that mean travel
distance increased by 12.7 miles if all procedures were performed at high-volume
hospitals. Most of these articles report small increases in travel distance
or travel time; however, our study showed a decrease in travel distance for
the majority of patients undergoing PTCA, a more common procedure.
Our analysis has some limitations. The distance calculations were made
using straight-line distances rather than travel route distances or travel
times. However, Grumbach et al17 report that
correlation between straight-line distances and travel time is 0.987 in New
York, except in New York City and Long Island. Also, under the minimum volume
standards, we did not take into account that patients living along state borders
may have been closest to hospitals in neighboring states, thereby further
reducing travel distances. Furthermore, the travel distances may also be inaccurate
because we measured distance from patients’ home addresses, although
some patients requiring urgent or emergent PTCA may not have been at home
at the time of the precipitating event. Using geographic centroids rather
than population centroids to approximate patient and hospital locations may
have introduced inaccuracies in the distance calculations. Also, our analysis
assumes that all patients would make the most efficient choice and select
the closest eligible hospital-operator if minimum volume standards were implemented.
The absence of clinical detail is an inherent limitation of administrative
databases and makes risk adjustment imprecise. Although we excluded transfer
patients from our mortality model, any patient who was discharged from one
facility and readmitted to another would be included in the analysis, thus
artificially depressing mortality estimates for the first facility. Similarly,
the data set does not include postdischarge mortality, so the mortality estimates
reflect in-hospital mortality only. Finally, the 3 states in our analysis
have certificate-of-need laws. It is possible that states without certificate-of-need
laws for angioplasty facilities may have more low-volume hospital-operator
pairs, which would result in more patients traveling farther to reach a hospital-operator
and meeting minimum volume requirements.
Proponents of regionalization often cite cost savings for payers and
mortality differences between high-volume and low-volume hospital-operator
pairs as the main advantages.17 If high-volume
hospital-operator pairs have lower costs and mortality rates, it would be
in payers’ interests to encourage patients to seek care from those hospital-operator
pairs. Report cards documenting quality may help to identify high-volume hospital-operator
pairs; however, payers must actively use quality data in contracting arrangements
to affect quality of care.19
The potential consequences of regionalization are both positive and
negative. Ceasing to provide one service may detract from other services at
the same hospital.11 For example, removing
interventional cardiologists from a low-volume hospital may cause patients
admitted for other noninterventional cardiac diagnoses to not receive high-quality
care due to the hospital’s inability to attract and maintain an experienced
cardiology staff. Alternatively, and contrary to the hypothesis of our study,
low-volume hospital-operator pairs could precipitate a volume shift and avoid
exiting the market by drawing procedures from nearby high-volume hospital-operator
pairs that perform well above minimum volume standards. Available studies
suggest that some degrees of regionalization can be achieved with minor consequences
for patient travel. However, potential costs of regionalization not related
to travel must be examined before such policies can be recommended.
Corresponding Author: Kevin A. Schulman,
MD, Center for Clinical and Genetic Economics, Duke Clinical Research Institute,
PO Box 17969, Durham, NC 27715 (firstname.lastname@example.org).
Author Contributions: Dr Schulman had full
access to all of the data in the study and takes responsibility for the integrity
of the data and the accuracy of the data analysis.
Study concept and design: Kansagra, Curtis, Schulman.
Acquisition of data: Curtis, Schulman.
Analysis and interpretation of data: Kansagra, Curtis.
Drafting of the manuscript: Kansagra.
Critical revision of the manuscript for important
intellectual content: Kansagra, Curtis, Schulman.
Statistical analysis: Kansagra, Curtis.
Administrative, technical, or material support: Schulman.
Study supervision: Schulman.
Previous Presentation: Presented in part at
the Alpha Omega Alpha Symposium at the Duke University School of Medicine;
May 11, 2004; Durham, NC.
Acknowledgment: We thank Kevin J. Anstrom,
PhD, for statistical advice and Damon M. Seils, MA, for editorial assistance
and manuscript preparation.