Do family physicians use more resources for inpatient care at an academic medical center?

by

John R. Holman

A thesis submitted in partial fulfillment of the requirements for the degree of

                        Master of Public Health

University of Washington

1999

Program Authorized to Offer Degree: Extended Degree Program, Department of Health Services

Ó Copyright 1999

John R. Holman

University of Washington
Graduate School

This is to certify that I have examined this copy of a master’s thesis by

John R. Holman

and have found that it is complete and satisfactory in all respects,
and that any and all revisions required by the final
examining committee have been made.

Committee Members:

Tom wickizer

Nancy Stevens

Joseph Yetter

Date: July 12 , 1999

Master's Thesis

In presenting this thesis in partial fulfillment of the requirements for a Master's degree at the University of Washington, I agree that the Library shall make its copies freely available for inspection. I further agree that extensive copying of this thesis is allowable only for scholarly purposes, consistent with "fair use" as prescribed in the U.S. Copyright Law. Any other reproduction for any purposes or by any means shall not be allowed without my written permission.

Signature

Date

University of Washington

Abstract

Do family physicians use more resources for inpatient care at an academic medical center?

by John R. Holman

Chairperson of the Supervisory Committee: Professor Tom Wickizer,

Health Services, Community Medicine

Introduction

Family physicians practice and train at community hospitals and academic medical centers. The rapid growth in the cost of medical care requires a careful appraisal of how resources are used in the care of patients in various settings. Family physicians use fewer resources in the care of patients in the inpatient and outpatient setting when compared to other primary care providers and specialists with no difference in outcomes. The use of resources by family physicians in different inpatient settings has not been explored.

Methods

A retrospective cohort design was used. Inpatient discharge data over a three-year period from a military community hospital and a military academic medical center were collected and analyzed. The military setting allowed comparisons to be made without the possibility of physician-induced demand for services. The facilities were compared with respect to length of stay, laboratory tests, radiographic tests and mortality rate. A subset of 196 non-obstetric charts was reviewed for consultation rate. Simple descriptive statistics, chi-square, t-test, and Mann-Whitney U tests were used. Linear and logistic regression were used to control for patient demographics and illness severity.

Results

The academic medical center had increased length of stay, use of diagnostic testing and consultations based on univariate analysis. However, after controlling for patient demographics and illness severity, use of consultation and laboratory testing were higher for the academic center. There was no statistically significant difference in mortality rate between the two institutions.

Conclusion

In this setting, family physicians at an academic medical center used more laboratory testing and requested consultations more frequently than family physicians practicing in a community hospital, controlling for patient demographics and illness severity. There may be a culture at the academic center that encourages increased use of resources for educational purposes. Physicians who are more "risk-averse" may prefer practicing in this setting, as they tend to order more testing and consultation. This information can help residencies as they affiliate with hospitals to provide patient care and resident teaching and for medical students as they select residency training programs.

TABLE OF CONTENTS

List of TABLES *

LIST of Abbreviations *

INTRODUCTION *

CHAPTER 1: Review of the Literature *

Resource Utilization *

Practice Setting *

Family Physicians *

Chapter 2: Methodology *

Hypothesis *

Study Design *

Study Population *

Measurement and Data *

Independent variables *

Data collection *

Data Analysis *

Chapter 3: Results *

Chapter 4: Discussion *

Bibliography 36

List of TABLES

Number Page

Table 1-Demographic Data *

Table 2 - Outcome variables *

Table 3-Factors Related to Length of Stay in Military Academic and Community Hospitals *

Table 4 -Factors Related to Lab Testing in Military Academic and Community Hospitals *

Table 5- Factors Related to X-ray Testing in Military Academic and Community Hospitals *

Table 6-Factors Related to Odds of Consultation in Military Academic and Community Hospitals *

Table 7-Factors Related to Odds of Formal Consultation in Military Academic and Community Hospitals *

Table 8-Factors Related to Odds of Informal Consultation in Military Academic and Community Hospitals *

LIST of Abbreviations

HMO. Health Maintenance Organization

COPD. Chronic obstructive pulmonary disease

1600. Major diagnostic codes

DRG. Diagnosis-related group

SIDR. Standardized inpatient data record

CHCS. Composite health care system

SPSS. Statistical package for social sciences

MAMC. Madigan Army Medical Center

NHB. Naval Hospital Bremerton

OR. Odds ratio

Acknowledgments

The author wishes to thank Joe Miller and Janet Schertzer of the Managed Care Division of Madigan Army Medical Center for their assistance in obtaining the data.

Dedication

The author wishes to dedicate this thesis to my family, Mary Ellen, Claire, Robert, Emily, Sara and Connor who had to live with me through this.

INTRODUCTION

Family physicians practice at community hospitals and academic medical centers. While there are pros and cons to each site, the rapid growth in the cost of medical care requires a careful appraisal of how money is spent in caring for patients in various settings. The utilization of resources for inpatient hospital care represents a significant portion of health care expenditures in the United States. While the costs of medical care have slowed in their increase, they are still much higher than other industrialized countries such as Canada or Great Britain. Family physicians may utilize resources at a lower rate than other providers with similar outcomes. At present there are no published studies comparing resource utilization for inpatient care among family physicians in different practice settings. Academic medical centers may have an environment that encourages increased use of resources for inpatient care. The objectives of the study are to compare length of stay, utilization of resources (laboratory and radiographic tests) and specialist consultation rates for family practice inpatients at a community hospital (Naval Hospital, Bremerton, Washington) versus an academic medical center (Madigan Army Medical Center, Tacoma, Washington). Both hospitals are military facilities in the Puget Sound area caring for active duty and retired families and have family practice residency training programs of the same size.

CHAPTER 1: Review of the Literature

Resource Utilization

While physician salaries account for a small amount of health care costs, medical doctors influence, directly or indirectly, 70 percent of all health care expenditures. A significant portion of health care costs is a result of inpatient medical care. The use of resources such as laboratory testing, specialist consultations, diagnostic imaging and other diagnostic tests and procedures may add significantly to the cost of inpatient care. More importantly, the utilization of these resources can result in increased lengths of stay, more complications and more surgeries. ^, Both clinical factors and physician specialty affect the ordering of diagnostic testing. ^, It is not just the sicker patients who have more diagnostic testing performed. The ordering of lab tests can also be very inefficient. One study indicated that 30 percent of all laboratory tests had no impact on patient care. Another evaluation of potentially avoidable laboratory use found 25 to 34 percent of lab tests were felt to be "inappropriate".

Of all the resources, specialist consultation has the greatest effect on resource consumption. This occurs due to increased use of laboratory testing, diagnostic imaging and other diagnostic testing by the specialists and the increased length of stay required to complete the tests and procedures. ^{4, ,} In a 1990 study of diagnostic test use by specialist consultants, patients who had consultation had an increase in stay of over 8 days and an increase in charges of over $6000 (1990 dollars). The consultants tended to order invasive, technologically sophisticated and expensive tests that were within their specialty. In fact, consultants recommended 35 percent of all diagnostic tests performed. There was no difference in in-hospital mortality between patients who had consultation and those who did not. Patients who received consultation were definitely sicker, but when stratified by severity of illness, patients in each group who had consultations had a longer length of stay.

Why do attending physicians request consultations? The reasons include the need for specialized procedures, the need for expert advice when dealing with uncertainty in the diagnosis and management of the patient’s illness, and, in the academic medical center, teaching. ^{, ,} Pressure from the consultants may influence attending physicians to order tests that may be "inappropriate" or "excessive". A physician’s training, experience or specialty may influence their need for consultation. ^, The ritual of requesting specialty consultation and the overutilization of diagnostic testing and procedures may be contributing to the rising costs of inpatient medical care.

Practice Setting

Academic medical centers are the hub of medical teaching. They are often affiliated with a university medical school and have the latest in medical technology. The medical center is populated by medical and surgical subspecialties and, until recently, avoided by primary care specialists. Teaching hospitals may use more resources than community hospitals. In 1977, medical teaching units ordered 50 percent more laboratory tests and had double the consultation rate of non-teaching units. Teaching hospitals are one-third more costly than nonteaching hospitals (adjusted for diagnosis related group case-mix). Family practice residents who practiced at a university center ordered more tests for "academic reasons" than residents who admitted to a community hospital. In an interesting study in 1994, 36 interns in an internal medicine program completed questionnaires on caring, medical knowledge and clinical judgement. Their lab utilization scores were calculated from hospital data and adjusted by severity of illness. The analysis indicated that increased clinical judgement was related to less laboratory utilization, while increased medical knowledge was associated with more laboratory utilization. The presence of large numbers of subspecialists at academic teaching centers with their depth of knowledge of their specialty may foster increased laboratory utilization. Other factors in the environment of the teaching hospital, such as aversion to risk, resident teaching or clinical research, may also contribute to increased utilization of resources.

A recent regional study questions this belief. A retrospective cohort study in northeast Ohio examined the severity-adjusted mortality and length of stay in major teaching hospitals, minor teaching hospitals and nonteaching hospitals. Both mortality and length of stay were lower for teaching hospitals than for nonteaching hospitals. Minor teaching hospitals, those institutions with only one residency training program, had no change in the length of stay. However, a higher mortality rate was detected.

The utilization of resources at academic medical centers is typically greater than for community hospitals. Previously, this has been associated with no change in clinical outcomes. However, recent studies suggest that in some regions, teaching hospitals may improve patient survival with no change or a reduction in resource utilization, measured in length of hospital stay.

Family Physicians

Family physicians have been compared to other primary care specialists with regard to their resource utilization for outpatient care. In a study from 1980, family physicians used fewer diagnostic tests than internists and general internists use fewer diagnostic tests than subspecialists in similar clinical situations . Similar studies in 1983 and 1984 reproduced these results ^, . In the care of chronic obstructive lung disease, similar results were found in a 1987 article. In a report from the Medial Outcomes Study, there were no consistent advantages for any specialty in the care of hypertension and non-insulin-dependent diabetes mellitus. Family physicians and internists may underutilize angiotensin converting enzyme inhibitors in the treatment of congestive heart failure. A study from a large health maintenance organization (HMO) showed asthma patients cared for by allergists had improved health function status scores on the SF-36 scale compared to those cared for by generalists. Family physicians are more parsimonious in their utilization of diagnostic testing than internists and subspecialists in the ambulatory setting with similar outcomes, although two recent studies question this. ^{28, 29}

There are no published studies comparing the use of resources by family physicians in caring for inpatients at an academic medical center compared with a community hospital. The current literature compares the use of inpatient resources by family physicians and internal medicine providers. In the treatment of diabetic ketoacidosis, internists had longer hospitalizations (5.1 vs. 4.6 days) and increased lab usage with no difference mortality or morbidity. ³ In critical care, family physicians and general internists showed no difference in resource utilization, death rate, readmission rate, median hospital charges or length of stay with adjustment for severity of illness. For the two specialties, the characteristics of their patients contributed more to explaining variations in the length of stay than did interspecialty differences. In a cohort trial from 1989, the average cost of hospitalization was 25 percent higher for internal medicine patients and they had a length of stay 19 percent longer. The major influence on these variables was the care received from the specialists who consulted on these patients. ⁴

The involvement of subspecialists may affect resource utilization as well as mortality and morbidity. A recent article on the inpatient care of patients with congestive heart failure showed improved clinical outcome (lower risk of readmission for congestive heart failure within 6 months) when a cardiologist was involved in the care. The patients also had longer hospital stays and increased diagnostic testing. ⁶ In contrast, patients admitted for severe exacerbation of their chronic obstructive pulmonary disease (COPD) had no improved outcome and no improved survival when cared for by a pulmonologist than when cared for by a generalist. The practice characteristic that may influence hospital charges most is not what services the primary care provider gives, but rather what referrals they make.

Resource utilization, in the form of diagnostic testing and specialist consultation, is an important contributor to the cost of medical care. Up to 30 percent of testing may be "unnecessary". The setting of care, the academic center or the community hospital, may also be an important contributing factor. Academic medical centers may be more costly than community hospitals in caring for similar patients. Finally, the specialty of the health care provider may also influence the utilization of resources for inpatient care. Family physicians seem to provide care using fewer resources than other primary care providers or specialists do.

Chapter 2: Methodology

Hypothesis

Family physicians who care for inpatients in a community hospital-based environment will have a shorter length of stay (in days), use fewer laboratory tests, diagnostic imaging, and specialist consultations than family physicians who care for inpatients in an academic medical center, controlling for patient demographics, severity of illness, and major diagnostic category of discharge diagnosis-related group. This hypothesis reflects the notion that the culture of an academic medical center encourages increased length of stay, diagnostic testing and specialist consultations. This culture, with it’s large population of subspecialists, clinical research and focus on medical training, promotes longer stays with increased use of resources as more tests and procedures are done to educate learners about the disease process.

Study Design

This study represents a retrospective cohort design. Data were retrospectively obtained for three years. Two groups of subjects were identified; those patients cared for by family physicians and discharged from Bremerton and those cared for by family physicians and discharged from Madigan. This was the key independent variable in the study. The most important dependent variable evaluated was length of stay. Other dependent variables analyzed include number of laboratory tests, number of diagnostic imaging tests and number of consultations performed. Other covariates entered in the analysis included age, sex, ethnicity, severity of illness and patient’s diagnosis. Both facilities were assumed to be equally efficient in making discharge arrangements for their patients. The quality of faculty and trainees was also assumed to be equal.

Study Population

The study population consists of all patients consecutively discharged from the family practice service during calendar years 1994, 1995 and 1996 from Bremerton Naval Hospital and Madigan Army Medical Center. The entire year was selected to avoid any bias due to increasing experience of house staff during their training year. If patients received the majority of their care from another service such as the intensive care team or cardiology, they were excluded. Patients admitted to the intensive care unit (ICU) at Bremerton had their care provided by the family practice service. Approximately ten percent of monthly family practice admissions were to the ICU. Patients admitted to the intensive care unit at Madigan were cared for by the subspecialist intensive care team without involvement of the family practice service. Identification and exclusion of those patients admitted to the intensive care unit at Bremerton was impossible with our data source. A power analysis was performed to determine the number of subjects needed to detect a difference of 0.25 days in the length of stay assuming a standard deviation of 1.0 days using an a of 0.05 and a 1-b of 0.80. This analysis indicated that 251 subjects were adequate to detect this difference. The total number of subjects from Madigan was 2839 and from Bremerton, 2221. Given the study populations are greater than 251 per group, this study had more than adequate power to detect a change in length of stay. The same analysis was performed for consultation review. The number of subjects needed to detect a difference of 3 percent in the referral rate was calculated assuming a standard deviation of 2.5 percent using an a of 0.05 and a 1-b of 0.80. This analysis indicated that a total of 196 charts were necessary to detect the change. As discussed below, data for consultation was obtained by chart review. Based on the power analysis 196 charts were selected for review. While equal groups of 98 were attempted, unequal totals resulted after random computer selection. Ninety-two charts from Madigan were reviewed for consultation rate and 104 from Bremerton.

Measurement and Data

Dependent variables

The primary dependent variable was average length of stay. Three other dependent variables were included in the study; the numbers of laboratory tests, diagnostic imaging tests, and consultations obtained during the inpatient episode. Each variable is defined as follows:

• Length of stay: Number of days from admission to discharge (Continuous measure)
• Laboratory tests: Total number of test done on blood or body fluids (Continuous measure)
• Number of consults: Total number of requested consultations (Continuous measure)
• Number of diagnostic imaging tests: Total number of radiological procedures performed (Continuous measure)
• Mortalities and mortality rate: Total deaths during study period and total deaths divided by total discharges (Continuous measure)

Independent variables

The primary independent variable was hospital site, or location of care. Other covariates include patient demographics such as age, sex and ethnicity, DRG and MDC, and case-mix index, a measure of illness severity. These variables are defined as:

• Location of care: Patients either received their care at the Naval Hospital Bremerton (community hospital) or Madigan Army Medical Center (academic medical center). These facilities are within 45 minutes of one another, have family medicine residency training programs with the same number of residents, and practice full-spectrum family medicine. Since all providers were salaried, there was no economic incentive to order extra tests or consultations, removing the ability of the physician to induce demand. (Categorical or binary)
• Age: Patients’ age at the time of discharge (Continuous measure)
• Sex: Male or female (Categorical or binary)
• Ethnicity: Ethnic origin as described by the patient at the time of admission (Categorical)
• Diagnosis related grouping (DRG) and major diagnosis code (MDC): Clustering of inpatient diagnosis codes based on the organ system involved (Categorical)
• Case-mix index: A measure of illness severity calculated by the inpatient records department on all patients. The index uses age, gender, and number and type of diagnoses, procedures and co-existing conditions to determine a number that relates to the overall severity of illness. (Continuous measure)

Data collection

The standardized inpatient data record (SIDR) for each patient discharged from the family practice services of the two hospitals in calendar years 1994, 1995 and 1996 was obtained. The following data was transferred from the SIDR database and entered into a separate database:

• Demographics such as age, sex and ethnic status
• Length of stay
• DRG group and primary ICD-9 diagnosis
• Additional diagnoses or coexisting illnesses
• Diagnostic and therapeutic procedures
• Discharge status
• Case-mix index

The Composite Health Care System (CHCS), a computer database for laboratory, radiographic and pharmaceutics ordering and tracking, was used to determine the number of lab and x-ray tests completed for each patient during their hospitalization period. This data was also entered into the same database directly from the CHCS report.

A randomly selected subset of approximately 196 patient charts was reviewed to determine the consultation rate from each hospital. The charts were selected from the top 3 non-obstetric DRGs. There were two types of consultations noted in the chart. The most common was the formal consultation with a specialist that included a note in the chart and detailed recommendations. There were also a large number of "curbside" consultations. "Curbside" consults were determined by the name and recommendations of a specialist present in the inpatient chart without an order for consultation or a note by the specialist. Data from these two types of consultations were collected and analyzed separately as well as together.

Data Analysis

The data was exported into the Statistical Package for Social Sciences (SPSS) for analysis. The variables length of stay, laboratory and radiographic tests were log transformed because of skewness. Chi-square tests were performed on categorical variables. The t-test was used to compare the means for the two populations for the continuous, normal data. The Mann-Whitney U test was used to compare continuous variables where distribution did not meet parametric statistical assumptions. Since many patients did not receive any laboratory or radiographic testing, the data were recoded into categorical variables and analyzed with the chi-square test. Multiple linear regression was used to determine the factors that contributed to the patients’ length of stay and to adjust for the effects of location of care, age, sex, ethnicity, diagnostic category and case-mix index on the dependent variables. The general linear regression model is y_i = a + b₁x₁ + b₂x₂ + b₃x₃ +….…b_nx_n + e_I, where y_i is the dependent variable, a is the constant, x₁, x₂,…x_n are the independent variables or covariates, b₁, b₂…b_n are the estimated regression coefficients and e_i is the error term. The key dependent variable is length of stay. Other dependent variables analyzed are numbers of laboratory tests, radiographic tests and consultations. The key independent variable (x₁) was location of care, either Bremerton (coded as 0) or Madigan (coded as 1). A positive estimated regression coefficient for x₁ indicates increasing use of resources for patients at Madigan. Covariates include age, sex, ethnic status, diagnosis, and severity of illness. Logistic regression was performed to determine the effects of hospital type on consultations, adjusting for the same covariates. An adjusted odds ratio that is positive and significantly greater than 1.0 indicates patients at Madigan are more likely to have consultation. Statistical significance is set at a p value of 0.05 based on a two-tailed test.

Chapter 3: Results

Table 1 shows the basic demographic data by hospital. The patients at Madigan (MAMC) were older and more likely to be male. The fact that patients at Madigan are older can be explained by the finding than Madigan cares for a higher percentage of retired service members and their families. Bremerton (NHB) has a more diverse ethnicity of patients. Both hospitals cared for predominately white patients, however. Bremerton’s population was 89.8 percent white and Madigan’s was 94.2 percent. The five most common MDCs are similar between the hospitals, although Bremerton has a higher percentage of obstetrical admissions than Madigan.

Table 2 lists mean and median values for the outcome variables examined. Average length of stay was less for Bremerton compared with Madigan. Bremerton inpatients stayed 2.4 days compared with 2.8 days at Madigan. The mean number of laboratory and radiographic tests done are also lower for Bremerton. Inpatients at Bremerton had 4.7 labs and 0.42 radiographic tests performed while those at Madigan had 8.1 labs and 0.56 radiographic tests. The case-mix index is higher for Madigan, suggesting the treatment of more seriously ill patients. Consults were higher at Madigan as well with an average of 0.92 consults per patient compared with 0.80 at Bremerton.

A significant number of patients had no labs, x-rays or consults performed. For example, 43.8 percent of patients at Bremerton and 29.8 percent of the patients at Madigan had no labs done. Eighty-five percent of Bremerton inpatients and 81.1 percent of Madigan inpatients had no radiographic tests performed. The data for laboratory tests, radiographic tests and consultations were recoded into categorical variables representing different levels of resource use (Not shown in table). Laboratory tests were recoded as follows: 0 to 1 tests, 2 to 5 tests, 6 to 15 tests and greater than 15 tests. Chi-square analyses showed than more inpatients at Bremerton had 0 to 1 labs done during their admission (49.8 vs. 39.6 percent) than at Madigan. Bremerton patients more frequently had 2 to 5 labs done and patients at Madigan more often had over 6 labs done (29.2 percent vs. 16 percent, respectively, p <0.01). Analysis by subgroups (non-obstetrical MDCs, obstetrical MDCs, respiratory, circulatory and gastrointestinal MDC) showed only nonstatistically significant differences in the number of lab tests.

Radiographic tests were recoded as follows: 0 tests, 1 test, 2 to 3 tests and greater than 3 tests. Chi-square analyses of the radiographic tests showed that patients at Bremerton more often had no testing done (85 vs. 81 percent). Inpatients at Madigan more often had 1 test (7 percent vs. 9.4 percent) or 2 to 3 tests performed (4.3 vs. 5.7 percent). Slightly more patients at Madigan had more than 3 radiographic tests done (3. percent vs. 3.8 percent, p value is <0.01). Analysis by subgroups showed nonstatistically significant differences among groups.

Consultation data were recoded as follows: 0 consults, 1 to 2 consults and greater than 2 consults. Chi-square analyses for the subset of subjects that had chart reviews for consultations found that inpatients at Bremerton were more likely (p <0.01) to have had 0 consults (55 vs. 37 percent) or more than 2 consults performed (9.6 vs. 2.2 percent). Inpatients at Madigan more likely to have had 1 to 2 consults performed (61 vs. 36 percent).

Table 3 presents multiple linear regression results for the key dependent variable, log-transformed length of stay. The key independent variable was hospital. Covariates for all tables included age, sex, ethnicity, case-mix index, and the top 5 MDCs. The reference groups were white ethnic status and newborn MDC. The adjusted R² is 0.20. The variable representing hospital was not independently associated with length of stay, estimated regression coefficient –0.007, p = 0.68. Some of the other covariates that independently contributed to this model were age, case-mix index and MDC.

Tables 4 through 8 display results of other dependent variables analyzed in this study. In particular, table 4 shows the results of the regression model for the dependent variable log-transformed laboratory tests. The independent variable, covariates and reference groups are as previously noted. The reference groups were white ethnic status and newborn MDC. The adjusted R² is 0.31. The estimated regression coefficient for the variable representing hospital is 0.18 (0.07 to 0.28, 95 % CI, p<0.01), which corresponds to 1.2 more lab tests performed for each patient at Madigan compared to Bremerton. The covariates that were also related to laboratory testing are age, case-mix index, MDC, and length of stay.

Table 5 displays the results of the regression model for the dependent variable log-transformed radiographic tests. The independent variable, covariates and reference groups are as previously noted. The adjusted R² is 0.38. Contrary to the hypothesis, patients at Bremerton had more radiographic tests performed. The estimated regression coefficient for the variable representing hospital was –0.19 (-0.24 to -0.13, 95 % CI, p<0.01), which corresponds to 1.2 more radiographic tests for the average patient at Bremerton. The covariates that were related to radiographic testing were age, case-mix index, MDC, length of stay, and having black or unknown ethnic status. Blacks had 1.3 fewer radiographic tests performed compared to whites.

Tables 6 through 8 report logistic regression analysis for total, "curbside", and formal consultation. The dependent variable is whether or not a consultation was obtained. The independent variable for all models was hospital. The covariates for all models were length of stay, ethnic status, sex, age, and case-mix index. The reference groups were white ethnic status and newborn MDC. Table 6 presents the logistic regression analysis for combined formal and curbside consultation. Patients cared for by family physicians at Madigan had over double the odds of having a consult than those cared for at Bremerton (adjusted OR = 2.12, p=0.03). Covariates that were independently related to consultations were CMI and length of stay.

Table 7 shows the logistic regression analysis for formal consultation. The dependent variable is whether or not a formal consultation was obtained. Patients cared for by family physicians at Madigan had nearly double the odds of having a consult than those cared for at Bremerton, but this finding was not statistically significant (adjusted OR = 1.82, p = 0.09). This odds ratio is similar to that obtained with combined consultations. With a larger sample size, statistical significance would likely be reached. Covariates that were independently related to formal consultation were CMI and length of stay.

Table 8 displays the logistic regression analysis for "curbside" or informal consultation. The dependent variable is whether or not an informal consultation was obtained. Patients cared for by family physicians at Madigan had double the odds of having a consult than those cared for at Bremerton, but this finding was not statistically significant (adjusted OR = 2.00, p = 0.11). This odds ratio is similar to that obtained with combined consultations. With a larger sample size, statistical significance would likely be reached. Covariates that were independently related to informal consultation were age and female gender.

Table 1-Demographic Data

		NHB	MAMC	p
Age		20.5 + 22.9	29.9 + 27.5	<0.01
Sex	Male	64.3 %	68.9 %	<0.01
	Female	35.7 %	31.1 %	<0.01
Ethnicity	White	89.8 %	94.2 %	<0.01
	Non-White	10.2 %	5.8 %	<0.01
MDC	Obstetric	40.8 %	30.4 %	<0.01
	Newborn	37.6 %	25.3 %	<0.01
	Circulatory	5.3 %	10.3 %	<0.01
	Respiratory	4.4 %	8.1 %	<0.01
	Gastrointestinal	3.1 %	5.2 %	<0.01
Category	Active Duty	86 %	69.7 %	<0.01
	Retired	11.1 %	25.1 %	<0.01
	Other	2.9 %	4.7 %	<0.01
Discharge Disposition	Home	96.2 %	98.1 %	<0.01
	Transfer	2.1 %	0.9 %	<0.01
	Died	0.7 %	0.8 %	NS

Table 2 - Outcome variables

	NHB		MAMC
	Mean	Median	Mean	Median	p
Length of stay	2.4 + 2.1	2.0	2.8 + 2.9	2.0	<0.01
Laboratories	4.7 + 11.4	2.0	8.1 + 19.2	3.0	<0.01
Radiographics	0.42 + 1.7	0	0.56 + 2.3	0	<0.01
Total Consults	0.80 + 1.2	0	0.92 + .90	1	0.04
Formal Consults	0.41 + 0.75	0	0.55 + 0.72	0	0.06
"Curbside" Consults	0.38 + 0.4	0	0.37 + 0.45	0	NS
Case-mix index	0.46 + 0.65	0.39	0.60 + 0.56	0.47	<0.01

Table 3-Factors Related to Length of Stay in Military Academic and Community Hospitals

	Estimated Regression Coefficient	p	95 % Confidence Interval
Constant	0.49	<0.01	0.45	0.53
Hospital	-0.007	0.68	-0.04	0.03
Age	0.006	<0.01	0.005	0.007
Sex	-0.012	0.57	-0.05	0.03
Case-mix index	0.36	<0.01	0.33	0.39
Respiratory MDC	-0.10	0.01	-0.18	-0.02
Circulatory MDC	-0.32	<0.01	-0.41	-0.24
Gastrointestinal MDC	-0.11	0.02	-0.20	-0.02
Obstetric MDC	-0.16	<0.01	-0.21	-0.11
All other MDCs	-0.004	0.24	-0.11	0.03
Asian/Pac Islander	0.03	0.62	-0.08	0.13
Black	-0.06	0.35	-0.18	0.06
Hispanic	-0.006	0.91	-0.11	0.10
SE Asian	0.11	0.17	-0.05	0.26
Unknown	-1.16	<0.01	-1.59	-0.72

Table 4 -Factors Related to Lab Testing in Military Academic and Community Hospitals

	Estimated Regression Coefficient	p	95 % Confidence Interval
Constant	-0.80	<0.01	-0.92	-0.68
Hospital	0.18	<0.01	0.07	0.28
Age	0.007	<0.01	0.004	0.010
Sex	-0.009	0.88	-0.13	0.11
Case-mix index	0.13	<0.01	0.035	0.23
Length of Stay	0.18	<0.01	0.16	0.20
Respiratory MDC	0.93	<0.01	0.68	1.18
Circulatory MDC	1.7	<0.01	1.41	1.96
Gastrointestinal MDC	1.3	<0.01	1.02	1.58
Obstetric MDC	-0.56	<0.01	-0.72	-0.40
All other MDCs	0.95	<0.01	0.75	1.15
Asian/Pac Islander	-0.02	0.92	-0.33	0.29
Black	0.34	0.08	-0.04	0.71
Hispanic	0.02	0.89	-0.29	0.34
SE Asian	0.02	0.92	-0.45	0.50
Unknown	-0.56	0.41	-1.9	0.77

Table 5- Factors Related to X-ray Testing in Military Academic and Community Hospitals

	Estimated Regression Coefficient	p	95 % Confidence Interval
Constant	-2.4	<0.01	-2.5	-2.3
Hospital	-0.19	<0.01	-0.24	-0.13
Age	0.005	<0.01	0.003	0.007
Sex	-0.02	0.60	-0.08	0.05
Case-mix index	0.19	<0.01	0.14	0.25
Length of Stay	0.14	<0.01	0.13	0.15
Respiratory MDC	0.85	<0.01	0.72	0.99
Circulatory MDC	0.59	<0.01	0.44	0.73
Gastrointestinal MDC	0.60	<0.01	0.45	0.75
Obstetric MDC	-0.28	<0.01	-0.36	-0.19
All other MDCs	0.55	<0.01	0.45	0.66
Asian/Pac Islander	0.02	0.84	-0.15	0.18
Black	-0.26	0.01	-0.46	-0.06
Hispanic	-0.10	0.23	-0.27	0.07
SE Asian	0.08	0.54	-0.17	0.33
Unknown	-0.91	0.01	-1.6	-0.2

Table 6-Factors Related to Odds of Consultation in Military Academic and Community Hospitals

	Adjusted Odds Ratio	p	95 % Confidence Interval for Adjusted Odds Ratio
Constant	N/A	<0.01	N/A	N/A
Hospital	2.12	0.03	1.08	4.20
Age	1.009	0.15	0.99	1.02
Sex	0.64	0.19	0.33	1.25
Case-mix index	3.59	0.04	1.08	11.93
Length of stay	1.49	<0.01	1.18	1.88
Asian/ Pac Islander	0.47	0.53	0.05	4.85
Black	0.28	0.13	0.05	1.48
Hispanic	0.42	0.78	0.00	4.27E10^21

Table 7-Factors Related to Odds of Formal Consultation in Military Academic and Community Hospitals

	Adjusted Odds Ratio	p	95 % Confidence Interval for Adjusted Odds Ratio
Constant	N/A	<0.01	N/A	N/A
Hospital	1.82	0.09	0.92	3.63
Age	1.0007	0.91	0.99	1.01
Sex	0.63	0.19	0.32	1.27
Case-mix index	3.70	0.02	1.27	10.76
Length of stay	1.34	<0.01	1.12	1.62
Asian/ Pac Islander	0.002	0.70	0.008	1.07E10^11
Black	0.22	0.11	0.033	1.39
Hispanic	0.001	0.86	0.000	2.11E10^28

Table 8-Factors Related to Odds of Informal Consultation in Military Academic and Community Hospitals

	Adjusted Odds Ratio	p	95 % Confidence Interval for Adjusted Odds Ratio
Constant	N/A	<0.01	N/A	N/A
Hospital	2.00	0.11	0.86	4.66
Age	1.02	0.04	1.001	1.04
Sex	0.39	0.04	0.16	0.94
Case-mix index	1.10	0.82	0.48	2.56
Length of stay	0.96	0.76	0.77	1.21
Asian/ Pac Islander	2.35	0.49	0.21	26.42
Black	0.50	0.55	0.05	4.62
Hispanic	2968.02	0.72	0.000	2.56E10^22

Chapter 4: Discussion

The purpose of this study was to evaluate the use of resources by family physicians for inpatient care at a military academic medical center compared with a military community hospital. Length of stay, laboratory testing, radiographic testing and specialist consultations were the dependent variables examined which represented resources used for medical care. The hypothesis was that resource use would be higher at the academic center due to factors that include a culture that encourages increased use of testing for educational purposes. The results showed mixed evidence of increased resource use. For the key dependent variable, length of stay, there was no significant difference by hospital type. However, differences were noted for other dependent variables studied. As predicted, both laboratory testing and consultations were higher for the academic center. Contrary to the hypothesis, radiographic testing was found to be higher for the community hospital.

The covariates that also contributed independently to length of stay, laboratory testing and radiographic testing included age, case-mix index, and type of MDC. This is not surprising, as older and sicker patients often need more intensive use of resources to diagnose and treat illnesses. Certain MDCs were also associated with increased resource use. For example, patients in the respiratory MDC had much greater use of radiographic testing than patients in the obstetric MDC. This finding also follows logically since many patients admitted under the respiratory MDC require radiographic studies for diagnosis and assessment of treatment. Obstetric patients admitted for routine vaginal delivery rarely have radiographic testing. Black patients were associated with lower use of radiographic services independent of hospital, MDC, age, sex and case-mix index. The explanation for this finding is unclear.

In analyzing consultation data, multiple logistic regression showed the type of hospital to be important for the total number of consultations. The odds ratio showed double the chances of having a consultation for inpatients at Madigan. Other important variables associated with the likelihood of having a consultation were length of stay and case-mix index. When the data were analyzed separately for formal and informal consultations, the type of hospital was no longer found to be statistically significant, but was of borderline significance (p = 0.10). With a larger sample size, statistical significance would have been achieved. For informal consultations only, age was a statistically significant predictor. Being a female patient was associated with lower odds of having an informal consultation.

For this three-year period of patient discharges from the two institutions, the academic medical center showed increased use of laboratory testing and consultations while the community hospital had increased use of radiographic testing, independent of severity of illness, patient demographics and discharge MDC. Since this study was performed at military facilities with salaried physicians, the potential for "physician-induced demand" of services such as laboratory or radiographic testing and consultations was prevented. There are no fiscal incentives for providers to increase use of these services. In fact, minimizing test ordering saves money for the hospital. However, no differences in length of stay were identified after multivariate analysis was performed.

Why did the academic center have more ordering of laboratory testing and consultations? One explanation may be less experienced staff and residents at the academic center. We did not control for physician experience, which has been shown to affect referral rates.¹⁶ All faculty at both facilities are board certified and all residents graduated from accredited allopathic or osteopathic schools. However, there may be differences in the house staff and faculty that was not accounted for in the analysis that may have affected referral rates and laboratory testing.

Another study has indicated that the ordering of tests and referrals is related to different attitudes regarding taking risks. Physicians who are more "risk-averse" tend to request more specialist consultations, prescribe more drugs and order more diagnostic testing. Perhaps the physicians practicing at Madigan are more risk-averse and willing to order laboratory tests and consultations. It may be that physicians who are risk-averse actively seek to practice in a setting that would allow them maximum opportunity for specialist referral and laboratory testing.

The differences may be due to the differences in the delivery of care at the two hospitals. Bremerton has a much larger obstetric practice than Madigan. These patients are healthy and require little intervention in terms of laboratory testing, radiographic services or consultation. This aspect was controlled for in the multivariate analysis. Family physicians at Bremerton care for patients in the intensive care unit (ICU). Those at Madigan do not admit ICU patients. ICU teams with a critical care physician as attending care for their patients. This factor was not controlled for in the multivariate analysis. ICU patients tend to be sicker and often receive more specialist consultations, laboratory testing and radiographic services. This may explain the increase in radiographic testing seen in the multivariate analysis for Bremerton. Patients cared for in the ICU may have required more radiographic testing. If patients cared for in the ICU at Bremerton were removed from the analysis, inpatients at Madigan may have even more laboratory testing and consultations evident. Unfortunately, this cannot be done using the SIDR database.

Patient expectations may also affect the ordering of tests and consultations. Patients at Madigan know that they are in an academic center and may expect a certain level of diagnostic effort for their illness. They are also aware that their hospital charges do not increase with increasing use of services. Specialist consultants are often present on the ward making themselves available for opinions. Patients know this and may come to expect a specialist consultation. Conversely, patients at Bremerton are aware that they are in a community hospital with fewer resources and fewer specialists available and may alter their expectations for care.

The illness the patient was admitted for also had an effect on the type and number of resources used in care. The variable MDC was always an independent predictor of length of stay and number of diagnostic tests performed. The MDCs incorporate a number of DRGs. Performing a similar analysis with a larger database would allow controlling for the individual DRG. This may allow a more accurate picture of which individual diagnoses are associated with increased resource utilization. This database was too small to permit controlling for individual DRGs. Further studies can focus on increasing the size of the database to allow comparison of individual DRGs and attempting to improve on the matching of programs to decrease the variability in delivery of care.

A number of studies have shown that providers at academic teaching hospitals practice a more resource intense medicine relating to diagnostic testing, specialist consultation and length of stay. ^{2, 9, 12, 13, 14, 15, 19, , ,} Perhaps a different culture exists at the academic medical center. This culture may encourage, either overtly or covertly, the increased use of diagnostic testing and specialist consultation for "educational purposes". ^{12, 13, 14} This type of culture may exist because there are larger numbers of specialists available for consultation. Specialty residency and fellowship training programs typically exist only at academic medical centers to obtain the volume of patients needed for adequate training. This high concentration of specialists and specialists-in-training may be a form of "induced demand". In order for the training programs to justify their existence, increased consultation must occur. In addition, the patients of primary care providers need to be accessible to the specialists to improve their education.

House officers in primary care often rotate with specialists-in-training on different services in the academic medical center. As they care for patients, camaraderie often develops. This friendship may influence the primary care house officer to request specialist consultation more quickly than if they did not have this relationship. The primary care house staff may feel an obligation to assist their colleagues in specialty care who are also in training. Their patients may or may not need the specialty care. However, their house staff colleagues in specialty training often benefit from the education they receive by assisting in the care of hospitalized patients.

Specialists may be an easy source to obtain answers to patient care questions. Rather than trying to perform a literature search, obtain the articles and critically read and interpret the medical literature, primary care providers may find it easier to consult a specialist colleague. The answer may not be the most current recommendations from the recent literature, but it was easily obtained.

Madigan also has a large collection of referral and practice guidelines. The guidelines were written by a committee of physicians dominated by subspecialists. Even though military hospitals are not fee for service, there are incentives for subspecialty departments to justify their existence by recommending consultation. There is the potential for these guidelines to advise specialist consultation at an earlier point in the evaluation and treatment of patients, which may result in increase consultation.

There have been many studies of family physician use of resources and consultation for inpatient and outpatient care compared with other primary care providers and specialty care providers. ^{3, 4, 5, 6, 23, 24, 25, 26, 27, 29, 30, 31, , , , ,} Most indicate that family physicians use fewer resources in terms of diagnostic testing and specialist consultation than the other providers do. This is the first study to address the resource utilization of family physicians in two distinct practice settings: the academic medical center and the community hospital. It was determined that there was more use of laboratory testing and consultative services at the academic center and more use of radiographic services at the community hospital. The type of hospital did not affect length of stay. Case-mix index, patient age and discharge diagnosis were also important predictor variables for resource utilization. There were no differences in mortality.

When insurance companies and health maintenance organizations (HMOs) assess the cost of care at hospitals, the effects of the teaching environment should be considered. The academic medical center may generate more laboratory tests and consultations than the community hospital, even for family physicians, who use the least amount of resources. Whether this translates into improved outcomes is open for debate. In this study, the length of stay was different but accounted for by the type of patient and their illness. The mortality rate for the two hospitals was not statistically significantly different. Future research should focus on identifying more subtle differences in patient outcomes between the two types of hospitals.

The findings of this study may be helpful for residency program directors as they affiliate with hospitals to provide inpatient training for their residents. Academic medical centers have certain advantages with regards to research, patient mix and training by specialists. The environment of the teaching hospital may also encourage more liberal use of certain resources. Whether or not this influences practice style after completion of residency training is unknown. Understanding the potential effects of the teaching hospital environment on resource utilization can help the faculty and program director prepare for a different type of inpatient care than is delivered in the community hospital. A residency program that is training family physicians to practice in community settings should carefully weigh the benefits and risks of inpatient training at an academic center. A future research question to be addressed is how the training environment ultimately influences practice patterns.

Medical students may also want to use this information as they select a family medicine residency program. Students who are more risk-averse may lean towards a program affiliated with an academic medical center. More complete availability of laboratory testing and consultative services can help these risk-averse students be more comfortable as providers. Students who are not as risk-averse may feel more comfortable at the community hospital where there is more freedom to care for the patient using clinical judgement rather than diagnostic testing and consultation.

Endnotes

1 Udall KS.  Cost efficiency: a new dimension of emphasis for family practice.  J Fam Pract 1984;19:601-603.

2 Stahl CJ, Fox RD.  The emergence of a medical school: impact on laboratory utilization and cost containment efforts in an affiliated teaching hospital.  Mil Med 1986;151:526-534.

3 Hamburger S, Barjnbruch P, Soffer.  Treatment of diabetic ketoacidosis by internists and family physicians: a comparative study.  J Fam Pract 1982;14:719-722.

4 Bertakis KA, Robbins JA.  Utilization of hospital services.  A comparison of internal medicine and family practice.  J Fam Pract 1989;28:91-96.

5 Carey TS, Garrett J. Patterns of ordering diagnostic tests for patients with acute low back pain. The North Carolina Back Pain Project. Ann Intern Med. 1996;125:807-14.

6 Reis SE, Holubkov R, Edmundowicz D, et al. Treatment of patients admitted to the hospital with congestive heart failure: specialty-related disparities in practice patterns and outcomes. J Am Coll Cardiol 1997; 30:733-8.

7 Hughes RA, Gertman PM, Anderson JJ et al.  The ancillary services review program in Massachusetts: experience of the 1982 pilot project.  JAMA 1984;252:1727-1731.

8 Gortmaker SL, Bickford AF, Mathewson HO, Dumbaugh K, Tirrell PC.  A successful experiment to reduce unnecessary laboratory use in a community hospital.  Med Care 1988;26:631-642.

9 Manu P, Schwartz SE.  Patterns of diagnostic testing in the academic setting: the influence of medical attendings subspecialty training.  Soc Sci Med 1983;17:1339-1342.

10 Selker HP, Beshansky JR, Pauker SG, Kassirer JP.  The epidemiology of delays in a teaching hospital.  The development and use of a tool that detects unnecessary hospital days.  Med Care 1989;27:112-129.

11 Braham RL, Ron A, Ruchlin HS, Hollenberg JP, Pompei P, Charlson ME.  Diagnostic test restraint and the specialty consultation.  J Gen Intern Med 1990;5:95-103.

12 Bosk CL.  Occupational rituals in patient management.  N Engl J Med 1980;303:71-76.

13 Charlson ME, Cohen RP, Sears CL.  General medical consultation: lessons from a clinical service.  Am J Med 1983;75:121-128.

14 Lee T, Pappius EM, Goldman L. Impact of interphysician communication on the effectiveness of medical consultations.  Am J Med 1983;74:106-12.

15 Williams SV, Eisenberg JM, Pascale LA et al.  Physicians’ perceptions about unnecessary diagnostic testing.  Inquiry 1982;19:363-370.

16 Vehvilainen AT, Kumpusalo EA, Voutilainen SO, Takala JK. Does the doctors' professional experience reduce referral rates? Evidence from the Finnish referral study. Scand J Prim Health Care 1996; 14:13-20.

17 Bugter M-AM, Winkens RA, Grol RP, et al. Factors predicting differences among general practitioners in test ordering behaviour and in the response to feedback on test requests. Fam Prac 1996; 13:254-8.

18 Schroeder SA, O’Leary DS.  Differences in laboratory use and length of stay between university and community hospitals.  J Med Educ 1977;52:418-420.

19 Yarnold PR, Martin GJ, Fenglass J, Curry RH, Bryant FB, Lyons JS, Hughes RL.  First-year resident’s caring, medical knowledge and clinical judgement in relation to laboratory utilization.  Acad Med 1994;69:996-998.

20 Hueston WJ.  The influence of a family practice residency on the costs of inpatient diagnostic testing.  J Fam Pract 1986;23:559-563.

21 Weil TP.  Hub-and-spokes regionalization.  The teaching hospital should be the hub of its healthcare market.  Health Prog 1991;72:24-31.

22 Rosenthal GE, Harper DL, Quinn LM, Cooper GS. Severity-adjusted mortality and length of stay in teaching and nonteaching hospitals. Results of a regional study [see comments]. JAMA1997; 278:485-90.

23 Noren J, Frazier T, Altman I, Derozier J.  Ambulatory medical care: a comparison of internists and family-general practitioners.  N Engl J Med 1980;302:11-16.

24 Bennett MD, Applegate WB, Chilton LA, Skipper BJ, White RE.  Comparison of family medicine and internal medicine: charges for continuing ambulatory care.  Med Care 1983;21:830-839.

25 Greenwald HP, Peterson ML, Garrison LP et al.  Interspecialty variation in office based care.  Med Care 1984;22:14-29.

26 Bertakis KD, Robbins JA.  Gatekeeping in primary care: a comparison of internal medicine and family practice.  J Fam Pract 1987;24:305-309.

27 Greenfield S, Rogers W, Mangotich M, Carney MF, Tarlov AR.  Outcomes of patients with hypertension and non-insulin-dependent diabetes mellitus treated by different systems and specialties.  Results from the medical outcomes study.  JAMA 1995;274:1436-1444.

28 Chin MH, Friedmann PD, Cassel CK, Lang RM. Differences in generalist and specialist physicians' knowledge and use of angiotensin-converting enzyme inhibitors for congestive heart failure [see comments]. J of Gen Intern  Med  1997; 12:523-30.

29 Vollmer WM, M OH, Ettinger KM, et al. Specialty differences in the management of asthma. A cross-sectional assessment of allergists' patients and generalists' patients in a large HMO [see comments]. Arch Intern Med 1997; 157:1201-8.

30 Hainer BL, Lawler FH.  Comparison of critical care provided by family physicians and general internists.  JAMA 1988;260:354-358.

31 MacDowell NM, Black DM.  Inpatient resource use: a comparison of family medicine and internal medicine physicians.  J Fam Pract 1992;34:306-312.

32 Regueiro CR, Hamel MB, Davis RB, Desbiens N, Connors AF, Jr., Phillips RS. A comparison of generalist and pulmonologist care for patients hospitalized with severe chronic obstructive pulmonary disease: resource intensity, hospital costs, and survival. SUPPORT Investigators. Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatment. Am J Med1998; 105:366-72.

33 Hillman BJ, Joseph CA, Mabry MR, Sunshine JH, Kennedy SD, Noether M. Frequency and costs of diagnostic imaging in office practice-a comparison of self-referring and radiologist referring physicians. N Engl J Med 1990; 323:1604-1608.

34 Grol R, Whitfiled M, Maesener J, Mokkink H. Attitudes to risk taking in medical decision making among British, Dutch and Belgian general practitioners. Br J Gen Prac 1990; 40:134-6.

35 Stefanu C, Newman RG, Pate ML, Chassie MB, Anderson RJ. Severity of patient's illness: an important factor contribuitng to laboratory costs in a teaching hospital. J Med Educ1984; 59:598-600.

36 Boice JL, McGregor M. Effect of resident's use of laboratory tests on hospitals costs. Jf Med Educ 1983; 58:61-64.

37 Lewenstein SR, Iezzoni LI, Moskowits MA. Prospective payment for physician services.  Impact on medical consultation practices. JAMA 1985; 254:2632-2637.

38 Johnson JE, Pinholt EEM, Jenkins TR, Carpenter JL. Content of ambulatory internal medicine practice in an academic Army Medical center and an Army community hospital. Mil Med 1988; 153:21-25.

39 Strauss MJ, Conrad D, Logerfo JP, Hudson LD, Bergner M. Cost and outcome of care for patients with chronic obstructive lung disease.  Analysis by physician specialty. Med Care 1986; 24:915-924.

40 Udvarhelyi IS, Goldman L, Komaroff AL, Lee TH. Determinants of resource utilization for patients admitted for evaluation of acute chest pain. J Gen Intern Med 1992; 7:1-10.

41 Hueston WJ, Applegate JA, Mansfield CJ, King DE, McClaflin RR. Practice variations between family physicians and obstetricians in the management of low-risk pregnancies. J Fam Prac1995; 40:345-51.

42 Jollis JG, DeLong ER, Peterson ED, et al. Outcome of acute myocardial infarction according to the specialty of the admitting physician [see comments]. N Engl J Med 1996; 335:1880-7.

43 Nash IS, Corrato RR, Dlutowski MJ, JP OC, Nash DB. Generalist versus specialist care for acute myocardial infarction. Am J Cardiol 1999; 83:650-4.

Return to Resource Center | Original Papers