How patients select sites for non-emergent acute care

Material Information

How patients select sites for non-emergent acute care
Bourn, Scott S
Publication Date:
Physical Description:
xii, 236 leaves : ; 28 cm


Subjects / Keywords:
Physician services utilization -- Case studies -- Colorado -- Denver ( lcsh )
Hospitals -- Emergency services -- Utilization -- Case studies -- Colorado -- Denver ( lcsh )
Medical care -- Evaluation -- Case studies -- Colorado -- Denver ( lcsh )
Hospitals -- Emergency services -- Utilization ( fast )
Medical care -- Evaluation ( fast )
Physician services utilization ( fast )
Colorado -- Denver ( fast )
Case studies. ( fast )
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )
Case studies ( fast )


Includes bibliographical references (leaves 215-236).
General Note:
Department of Health and Behavioral Sciences
Statement of Responsibility:
by Scott S. Bourn.

Record Information

Source Institution:
|University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
436443689 ( OCLC )
LD1193.L566 2009d B68 ( lcc )

Full Text
Scott S. Bourn
Bachelor of Science in Nursing, University of Colorado, 1979
Master of Science in Nursing, University of Colorado, 1990
A thesis submitted to the
University of Colorado Denver
in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy, Health and Behavioral Sciences

This thesis for the Doctor of Philosophy
degree by
Scott Smith Bourn
has been approved
j Loo\
r Date

Bourn, Scott Smith (Ph.D., Health and Behavioral Sciences)
How Patients Select Sites for Non-Emergent Acute Care
Thesis directed by Professor Kitty Corbett
Emergency department (ED) utilization in the United States has dramatically
increased over the past 30 years. Most individuals seeking ED care have a primary
care provider and insurance. This study examined factors that insured individuals
consider when choosing where to seek care for non-emergent acute complaints, and
created and validated a decision tree model to predict where adults seek care.
Interviews were conducted with a convenience sample of 24 insured adults
who presented for care at an emergency department, urgent care center, and physician
practice in Denver, Colorado. Participants were 21-85 years of age, with diverse
ethnic, educational, and socioeconomic backgrounds. Interviews explored decision

factors to create an Ethnographic Decision Tree as described by Gladwin. Validation
of the model was based on a survey (N=27).
Interviews revealed four major themes related to the decisions to seek care
and choose a care site: Care Seeking, based on perceived severity and interruption in
activities of daily living; Access to primary care provider (PCP), based on satisfaction
with ability to get an appointment and willingness to see PCPs partners; Value of
Caregiver Relationships, based on the relative value of the therapeutic relationship vs.
time required to get an appointment; and Rules and Norms, concerning what
symptoms are appropriate to be seen in specific sites. These themes provided the
foundation for a decision model based on the following questions: Is this an
emergency?; How satisfied am I with prior PCP experience?; How confident am
I that I know what is wrong? Internal validity of the decision tree was 91%. The
decision model predicted 92% of the decisions of the survey respondents.
Patients who seek care for non-emergent acute conditions consider a variety
of factors related to symptom severity, prior PCP experience, and confidence of self-
diagnosis when choosing whether and where to seek care.
An understanding of how individuals choose where to seek care can guide
conversations with patients at the point of service, inform public policy consideration

of individuals who seek care in the wrong location, and lead to interventions that
honor the patient factors that drive care site selection.
This abstract accurately represents the content of the candidates thesis. I recommend
its publication.

Eight years ago it didnt really seem all that difficult to finish a doctorate
program or complete a small qualitative research study. But, as countless others who
have walked this path before can attest, it was. My successful completion is a
testimony to the help and support of many individuals.
I am indebted to the Medical Directors and staff of the three participating
facilities. You provided advice when asked, unending support, and access to the
patients with whom you are entrusted; I wish keeping your organizations anonymous
didnt prevent my publicly thanking you by name. My dissertation committee
deserves thanks for providing input, even when I didnt want it. Each of you has
shaped a part of this research, and of my understanding of the results, and you will
influence all the work 1 do in the future. I am also grateful to the HBS program
faculty who radically revised my thinking on the social aspects of patient behavior
and health care delivery.
This work eclipsed two employers and several job changes. Im grateful to
my co-workers for tolerating my mental and physical absences, and particularly to
Dighton Packard (you influenced my early thoughts on this topic more than youll
know) and Ron Thackery (who Im sure thought I only worked part-time there for a
while). Thanks to Lisa Rybowiak for protecting my calendar, and Sharon Ruiz for

reading, editing and cleaning up an otherwise haphazard document so it looked good
(and complied with the grad school rules!).
Several colleagues were invaluable in encouraging and supporting my quest,
and for reminding me that I would, in fact, finish. Among them Dr. Gregg Margolis
was a steady hand throughout, and soon-to-be Dr. Dave Williams provided a much
needed push at the end.
My greatest thanks go to my family who has blessed me with an unbelievable
measure of support. 1 am grateful to my children, most of whom started college
during this project, for your interest and encouragement, and to Chris for not bragging
too much when you graduated before me. Words are inadequate to convey thanks to
my wife, counselor, and best friend Marilyn. You gave the best gifts of all: time
(countless late nights and almost a whole year of Saturdays) and encouragement
(even when 1 didn't deserve it). I love you and promise not to return to school for at
least a few years.

CHAPTER 1. INTRODUCTION............................................1
CHAPTER 2. SPECIFIC AIMS...........................................4
CHAPTER 3. BACKGROUND AND SIGNIFICANCE.............................6
The Dynamics of Emergency Department Use.................6
Alternatives to the Emergency Department...............23
The Meaning and Interpretation of Symptoms.............30
Knowledge of and Access to Sources of Healthcare.........34
Selecting the Healthcare Site..........................38
The Role of Identity and Social Norms in Decision Making.49
CHAPTER 5. PRELIMINARY STUDIES....................................53
CHAPTER 6. RESEARCH DESIGN........................................56

Limitations of Quantitative Approaches....................56
Qualitative Methodology...................................59
CHAPTER 7. METHODS AND ANALYSIS......................................75
Settings and Sample.......................................75
Data Collection and Management............................81
Data Analysis.............................................83
CHAPTER 8. RESULTS...................................................93
The Interview Phase Sample Population.....................93
Qualitative Analysis.....................................105
CHAPTER 9. DISCUSSION..............................................160
Comparison with Previous Studies.........................160
CHAPTER 10. CONCLUSION..............................................175

CHAPTER 11. APPENDICES.................................179
CHAPTER 12. REFERENCES.................................215

8-1 Composite decision model with subject distribution..........................147
8-2. Composite decision tree model..............................................150
8-3. Model internal validity testing............................................151
8-4 Predicted distribution of survey subjects...................................153
8-5. Accuracy of decision points................................................154

8-1 Characteristics of Study Participants.......................................94
8-2 Unscheduled visits last 12 months...........................................97
8-3. Participant reasons for seeking care.......................................106
8-4 Codes related to access to PCP..............................................112
8-5 Relative Value of Relationships with Caregivers and Time....................117
8-6 Conditional and Unconditional Rules.........................................126

Each year there are approximately 115 million visits to hospital Emergency
Departments in the United States (Nawar, 2007). This represents a dramatic rise
estimated to be between 19% (Richardson, 2001b; Williams, 1996) and 45% (Taylor,
2001) over the past 10-15 years. This burgeoning use of the E.D. has created
widespread overcrowding and excessive delays in patient evaluation and treatment
(Derlet, Richards, & Kravitz, 2001; Reeder, 2001; Richardson, 2001b; Schull, 2001).
Severe overcrowding may even result in some patients leaving the E.D. without
receiving care even when they have significant medical problems that require urgent
or emergent intervention within days (Baker, 1991; A. B. Bindman, 1995; A. B.
Bindman, Grumbach, K, Keane, D, Rauch, L, Luce, JM, 1991). When E.D.
overcrowding results in extreme waiting times for patients, an unintended triage
process occurs that in some cases results in the inadvertent withholding of care to
patients who have urgent or emergent needs.
What kinds of conditions cause patients to seek care in the emergency
department? Many early critics of increasing E.D. utilization suggested that an
unacceptable number of non-urgent and non-emergent patients were using the E.D. as
a site for primary care (Gifford, 1980; Guterman, 1985; S. F. Hurley, Huggins, &
Jolley). There is widespread disagreement regarding the appropriateness of E.D.

utilization by health care consumers, in part due to difficulties in the definitions of
"urgent" and "emergent (Richardson, 2001 a). Rough estimates of E.D. patient
populations suggest that somewhere between 42% (Williams, 1996) and 53% (G.
Young, 1995) of patients presenting to the E.D. have an "urgent problem. Perhaps
the best study to date found that 68.4% of patients presenting to the E.D. had a
condition that required evaluation and treatment within 2 hours (Nourjah, 1997). In
the context of E.D. overcrowding, these data suggest that 30% 50% of the patients
presenting to the E.D. have non-emergent conditions that could be cared for in
another health care facility.
Is there a profile of patients who receive care in the E.D.? In some cases the
decision to visit the E.D. is made by factors and forces beyond the patients control.
Critically ill patients are transported to the E.D. by ambulance, and those who require
admission or re-admission to the hospital may also start by visiting the E.D. But what
about patients who dont have an urgent or emergent problem? The conventional
wisdom has been that patients seeking non-emergent care in the E.D. are either
uninsured or have no other access to health care (Cooke & Finneran, 1994). A
growing body of literature refutes this assertion. Users of the E.D. are not typically
poor, minority, or un-insured (J. Afilalo, et al., 2004; Lucas & Sanford, 1998; Ragin,
et al., 2005; M. A. Rubin & Bonnin, 1995; Shesser, Kirsch, Smith, & Hirsch, 1991;
Weber, Showstack, Hunt, Colby, & Callaham, 2005), and there appears to be no
difference in utilization of the E.D. between HMO members as compared to those

with other types of health care insurance, including Medicare/Medicaid (Reschovsky,
Patients will continue to seek care in the E.D. for a variety of conditions. To
prevent overcrowding and its associated patient care complications, one of two things
must happen. Either the E.D.'s capacity to treat non-emergent patients must increase
OR alternate sites for delivery of unscheduled acute care must be identified and
proven to consumers. To be successful, either of these options must be informed by
an understanding of the factors health care consumers consider when deciding where
to receive unscheduled, non-emergent care. That is the topic of this study.

This study will identify the factors that adults who have a choice consider
when deciding whether and where to seek unscheduled, non-emergent health care.
For the purposes of this study, consumers will be considered to have a choice if they
have some form of health insurance, including Medicare or Medicaid. In most
communities, consumers may select between the following care sites: emergency
department, urgent care clinic (typically staffed by 1-2 physicians without on-site
specialists or patient admission capability), or a physician office.
Specifically, the study will:
Aim # 1: Describe the factors and constraints that adults consider when
choosing where to seek care when they have an acute, non-traumatic medical
condition. This aim will be achieved through the use of in-depth qualitative
interviews of subjects who seek care in three different types of health care facilities as
identified above.
Aim #2: Create and validate a decision tree model that predicts where adults
will seek care based on the factors and constraints described in Aim #1. This aim will
be achieved through the analytical and Modeling process described by Gladwin and
others (L.C. Garro, 1998; C. H. Gladwin, 1976, 1989; C. Young, Garro, L, 1981).
The creation of a decision tree offers two advantages beyond the descriptive function
of Aim #1. First, it suggests how the factors identified in Aim #1 may be used to
better understand and influence the behavior of health consumers. Second, decision

trees are testable. Future validity tests of the decision tree will identify the accuracy
of the model in predicting consumer behavior in this and other populations. Without
the creation of a decision model that can be validated, factors and constraints remain
abstract concepts with little usefulness in establishing policy or guiding the decisions
made by consumers on who seek care for non-traumatic unscheduled health care.

The Dynamics of Emergency Department Use
By virtually all accounts, patient care activity in the emergency department
(ED) has dramatically increased since emergency medicine was recognized as a
specialty in the early 1970s. In 1993 a General Accounting Office report stated that
emergency department use had increased by 19 percent from 1985 to 1990. This
included a 34% increase in visits by Medicaid patients, a 29% increase in visits by
Medicare patients, and a 15% increase by uninsured individuals (Williams, 1996).
The trend has continued; according to the 2005 National Hospital Ambulatory
Medical Care Survey from 1995 through 2005 the annual number of ED visits
increased from 96.5 million to 115.3 million visits, a 20% increase. In addition to the
growing US population, this increase represents rising utilization, from 36.9 visits/
100 persons in 1995 to 39.6 visits/100 persons in 2005 (Nawar, 2007).
Complicating and contributing to increasing patient visits is the closure of
hospitals or emergency departments. In a study of emergency department visits
between 1988 and 1999 the number of emergency departments in the US decreased
from 5, 210 to 4,740a decrease of 9% (Taylor, 2001). These closures were largely
related to cost-reduction efforts by hospitals, which in turn were triggered by
economic forces related to managed care, as well as nursing shortages and the
downsizing of graduate education programs (Richardson, 2001b). The decline has
continued; between 1995 and 2005 the number of E.D.'s dropped from 4,176 to

3,795. driving the annual visits per ED from 23,119 to 30,388, an increase of 31 %
(Nawar, 2007). Increases in utilization are not uniform across the population. In a
study of utilization between 1993 and 2003 Roberts et al. found that increases in
visits were led by blacks (93%), and individuals between the ages of 65 and 74 (34%)
(Roberts, McKay, & Shaffer, 2008).
The E.D. as a Healthcare and "Social Safety Net
Why has there been such a dramatic increase in E.D. visits in the United
States? A number of studies suggest that the increase is related to major changes in
healthcare delivery. Two specific areas of interest are changes in in-patient capacity
and the social safety net. The increases in E.D. visits during the past two decades are
dwarfed by the reduction in in-patient capacity in Americas hospitals. According to
the American Hospital Association, there were 1,001,801 hospital beds in 5813
hospitals in 1981; that number had decreased to 829,000 hospital beds in 4,956
hospitals in 1991 (Association, 1999). By 2006 the number had further dropped to
802,658 beds in 4,927 hospitals, a reduction from 4.37 beds/1,000 people (1981) to
2.68 beds/1,000 people in 2006 (Health, 2008). Why has there been such a
remarkable decrease in available hospital beds? While the details behind this issue
are beyond the scope of this discussion, the simple answer is cost pressures. For the
past 25-30 years, medical care payers (including insurance companies and the
government) have been engaged in a series of efforts to decrease the cost of American
health care through careful control of healthcare utilization and access. The goal has

been to prevent unnecessary medical interventions and to assure that interventions
that were approved were performed in the most cost-effective location (Richardson,
2001a). One medical intervention that was carefully scrutinized was hospital
admission. As a result, a significantly lower number of patients have been admitted
to the hospital, or patients were admitted for a shorter period. American Hospital
Association statistics suggest that the rate of admission per 1,000 persons has
decreased from 158.8 in 1981 to 118.2 in 2006 (Health, 2008), and the National
Center for Health Statistics reported that the average hospital length of stay decreased
from an average of 6.7 days in 1990 to 5.6 days in 1995. When applied to the entire
US population, this represents a decrease of hospitalized days of 120.8 days/1,000
each yeara decrease of over 24 million days of hospitalization over the entire US
population {National Health Interview Survey, 1998). Because a high hospital census
is required for hospitals to remain financially solvent, decreasing admissions caused a
number of hospitals to reduce the number of available beds, or to close entirely. One
consequence of fewer and shorter hospital admissions is an increase in the number of
outpatient visits, including the emergency department.
What is the healthcare safety net? In a sweeping report entitled Americas
Health Care Safety Net: Intact but Endangered (Lewin, 2000) the Institute of
Medicine defined Americas health care safety net as a patchwork of hospitals,
clinics, financing and programs that serve as the default system to care for those
vulnerable populations who fall outside the economic and medical mainstreams.

One such vulnerable population is the growing number of uninsured individuals in the
United States. Despite an unprecedented period of economic prosperity that only
recently ended, the number of uninsured Americans totaled 42 million in the year
2000 (Asplin, 2001), and had increased to 47 million 15.8% of the total US
populationby 2006 (Census Bureau, 2007). This figure represents approximately
18% (1 out of 6) of non-elderly individuals in America (Adams, 2001). As might be
expected, the uninsured population is disproportionately made up of lower income
individuals. According to the Kaiser Family Foundation, 36% of families who make
less than $20,000 per annum are uninsured, as compared to only 11% of those making
greater than $80,000 each year (Health Insurance Coverage in America 2006, 2008).
The uninsured population is also disparate between regions of the country, individual
states, and racial/ethnic groups (Richardson, 2001b). In addition to the uninsured, the
under-insured population also continues to climb. According to the Institute of
Medicine, there were 44 million low-income individuals whose insurance was
inadequate to meet their needs, a group that includes Medicaid patients and
individuals with special medical needs (Richardson, 2001b). Based on these
statistics, over 35% of the US population is either un- or under- insured. These
individuals and families comprise the population served by the healthcare safety net.
The Emergency Department has been characterized as the safety net for the safety
net (Gordon, 2001) because of its 24 hour availability and its legal requirement
("Emergency Medical Treatment and Active Labor Act (EMTALA)," 1990) to

provide immediate care and stabilization to all individuals regardless of their ability
to pay (Cetta, Asplin, Fields, & Yeh. 2000; Pollock, 2001; Richardson, 2001a,
2001b). While the exact contribution the increasing number of "safety net patients"
has made to the increase in E.D. visits has not been determined, it has undoubtedly
played a major role in the stresses placed on shrinking emergency department
resources (Richardson, 2001a).
Changing Patient Acuity
The premise of many early critics of increasing E.D. utilization was that an
unacceptable number of non-urgent and non-emergent patients were using the E.D. as
a primary care site (Gifford, 1980; Guterman, 1985; R. Hurley, Freund, D, Taylor, D,
1989). Estimates regarding the percentage of inappropriate cases were reported to
have been as high as 50% (National Center for Health Statistics, 1994; Nonurgent
Use of Hospital Emergency Departments by Medicaid and Medicare Beneficiaries: a
Program Inspection for the Office of Inspector General, 1983). A 1990 Government
Accounting Office (GAO) study suggested that 43% of E.D. visits involved non-
urgent cases, with non-urgent defined as not life or limb threatening or did not
require immediate care and probably could have been treated in a doctors office or
clinic. The GAO source was an estimate of hospital officials, some of whom were
not directly involved in emergency care. A study evaluating 6,187 patients seen
during a 24 hour period in 56 U.S. hospital emergency departments concluded that
37% (range 23%- 72%) were non- urgent; the determination of urgency was based on

the triage criteria used by each individual hospital rather than a uniform scale (G.
Young, Wagner, MB, Kellermann, AL, Ellis, J, Bouley, D, 1994). In a 1996 study
Williams evaluated 24,010 E.D visits to six community hospitals in Michigan.
Thirty-two (32) percent of the cases were determined to be non-urgent", with
another 26% felt to be semi-urgent." Only 42% of the total caseload was felt to be
urgent" (Williams, 1996).
More contemporary studies have suggested that the severe reduction in in-
patient capacity and an increase in safety net activity has actually increased both the
frequency of E.D. visits and the acuity of the patients visiting the E.D., particularly in
the years following the institution of managed care (Adams, 2001). In the available
literature the quest to better understand patient acuity is somewhat flummoxed by the
absence of any standardized method of categorizing patients, or of a single statement
of what constitutes an appropriate or inappropriate E.D. patient (Richardson, 2001a).
However, a national probability survey (National Hospital Ambulatory Care survey)
was performed on 1997 data collected from 392 emergency departments. Actual
patient care records were reviewed and a uniform urgency criterion was applied to all
cases. Cases were considered to be emergent if the patient required evaluation
within 15 minutes of arrival in the E.D., urgent within 15-60 minutes, semi-
urgent within 1-2 hours, and non-urgent within 2-24 hours. In this study, 21% of
cases were classified as emergent, 32% as urgent, 15.4% as semi-urgent, 9.7% as
non-urgent, and 21.9% as unknown or no triage. Based on this landmark study,

only 9.7% were specifically identified as non-urgent. Conversely, 68.4% of cases
were deemed as requiring evaluation and treatment within 2 hours (Nourjah, 1997).
A more precise strategy for categorizing ED patient acuity and frequency of
utilization was used in a retrospective, cross-sectional study evaluating the clinical
and financial data for all ED patients (80,209) seen in a large urban teaching hospital
during 2001. One of the studys primary goals was to look for relationships between
individuals frequency of visits to the ED and their acuity in an effort to determine
whether frequent ED attenders were more likely to have non-emergent or non-urgent
problems. Patients were categorized by the triage nurse as being acuity A
(requiring resuscitation) through E (not urgent, could be referred elsewhere for
evaluation). Patients who were deemed to require evaluation and treatment within 30
minutes were categorized as A, B, or C. The results at first blush seemed to support
the notion that frequent ED attenders werent very sick. Individuals with only one
visit during the study period were most likely to be categorized as acuity A requiring
resuscitation (2.4%) and least likely to be categorized in acuity E (1.2%).
Conversely, individuals with more than 20 visits during the study period were least
likely to be categorized as acuity A (0.5%) and most likely to be categorized as acuity
E, non-urgent and able to be referred (6.1%). However, when all levels of utilization
were evaluated together a more interesting finding appeared; the percentage of
patients who were categorized at acuity levels A-C remained extremely stable at
86%-88% between patients with 1-20 visits during the year, and only dropped to 72%

for individuals who had over 20 visits during the study period. The authors
concluded Our data support the notion that the vast majority of frequent users of the
ED... have serious illness, being as likely as or more be triaged as
emergent and require hospitalization (Ruger, Lewis, & Richter, 2006).
Patient Factors Associated with E.D. Use
Speculation has abounded regarding the use of the E.D. as a de facto primary
care facility. A growing body of research has provided a sharper view of emergency
department users. A number of studies have debunked the pre-eminent belief that the
majority of individuals utilizing the E.D. for non-urgent conditions do so because of
an inability to pay (Cooke & Finneran, 1994). An early study evaluated 507
consecutive patients who presented at a major urban Level I trauma center (a very
high level center typically associated with a major teaching institution). Based on
their initial triage, patients were separated into urgent and non-urgent groups.
Following appropriate patient consent, the non-urgent patients (n=188) were asked
questions about their E.D. visit and previous utilization of the health care system.
Insurance/payment status was compared between the urgent and non-urgent groups.
Several trends were identified. Thirty-eight (38) percent of the non-urgent group had
commercial/private insurance, as compared to only 25% of the urgent group. There
was no statistically significant difference between the percentage of
Medicare/Medicaid patients between the urgent and non-urgent groups. Self-pay
(uninsured) patients were the most reluctant to use the E.D., with only 22% of the

non-urgent group uninsured as compared to 33% of the urgent group. The authors
suggested two hypotheses for the increased use of the E.D. by commercially insured
individuals. The first was that the E.D. was simply more convenient because of the
24-hour availability and the potential to get immediate diagnosis and treatment. An
additional theory was that insured individuals utilized the E.D. because
reimbursement by their insurer was more likely than if they utilized a private
physicians office (M. A. Rubin & Bonnin, 1995).
In another study, an observational case-control method was utilized to
compare the demographics, and socioeconomic status between 325 patients with
minor problems with 224 randomly selected patients from the same E.D. There were
no statistically significant differences in race, education, or socioeconomic status
between the two groups. The "minor group tended to have a lower frequency of
chronic illness and was more likely to have no established health care provider. The
rationale for choosing the E.D. was that it was easier and quicker to use than other
facilities (23.7%), no relationship with a health care providers (22.1%), unable to
make an timely appointment with regular health care provider (19%), referred by
private physician or employer (14.5%), problem perceived to be outside of expertise
of regular caregiver (10.2%), or away from home/regular caregiver (10.5%) (Shesser,
et al., 1991). A similar study compared the socioeconomic status, availability of
health care, insurance status, and reasons for E.D. visit of frequent E.D. users
(defined as having 2 E.D. visits within the last month) with the general E.D.

population. One hundred thirty four (134) frequent users were selected from a
general population of 6.523 E.D. patients. The "frequent users were more likely to
be black, and a higher percentage had Medicare or Medicaid than in the general E.D.
population. However, there was not a higher percentage of uninsured individuals,
and there was no difference in the availability of a primary care physician. The
"frequent user group had a significantly higher rate of admission to the hospital
(28%) as compared with the general E.D. population (16%). Membership in an HMO
was not different between the two groups. The frequent users identified the
following reasons for choosing the E.D.: thought problem was serious and needed
immediate attention (58%), referred by primary care physician (16%), couldnt wait
to see private physician (11%), have no physician (4%), convenience (3%), no
insurance, cant go elsewhere (2%), other (5%) (Lucas & Sanford, 1998). In a large
national survey of individuals with private insurance, Reschovsky et al. found that
there was no detectable difference in the use of inpatient care, emergency department
utilization, or surgery between individuals with HMO coverage as compared with
other types of conventional insurance. Apparently HMOs have not been successful in
"maintaining health and reducing the use of these expensive services (Reschovsky,
Several studies have evaluated use of the ED by elderly patients. A national,
population-based cross sectional study based on the 1993 Medicare Current
Beneficiary Survey evaluated 9,784 non-institutionalized patients over the age of 65.

Individuals over the age of 84 utilized the ED more frequently than the remainder of
the study population. Lower income, less education, and living alone were all found
to predict increased ED use. Frequent use was also inversely related to self-reported
health and ability to perform activities of daily living. There was no relationship with
having a regular source of medical care (Shah, Rathouz, & Chin, 2001). A more
recent study evaluated 4,310 individuals age 70 and above for patterns of ED use
during a four year period. A specific goal of the study was to determine the medical
necessity of the visits as measured by "high intensity" (visits that required highly
intensive medical therapy) and low intensity (visits that required little or no
intensive medical therapy). The majority of participants (56.6%) had no ED use
during the study period; 28.9% used the ED exclusively for high intensity visits, 5.7%
for exclusively low intensity visits, and 8.7% a mixture of high and low intensity
visits. The authors concluded that there are very few avoidable, low intensity ED
visits by older patients (Woiinsky, et al., 2008).
Two more recent studies have used the national Community Tracking Study
Household Surveys to evaluate ED utilization. The first studied ED utilization and
the characteristics individuals who used the ED during the 12 months of the study.
They found that 92% of the individuals who visited the ED during the study period
visited three or fewer times and accounted for 72% of all adult visits. The remaining
eight percent made four or more visits and accounted for 28% of all adults visits; for
the purposes of the study these individuals were considered frequent users. The

study results demonstrated that the vast majority of all ED patients were insured
(>85%) and reported having a usual source of care other than the ED (>87%). The
study found that frequent users were less likely to be uninsured or have no usual
source of care than less frequent users. Frequent users were also more than twice as
likely to have a family income below the poverty line and almost twice as likely to
report being in poor health. As a testimony to their poor health, 68% of the frequent
users reported 5 or more outpatient visits during the previous 12 months as compared
to only 39% of the less frequent users (Hunt, Weber, Showstack, Colby, & Callaham,
2006). The second study evaluated the impact of uninsured individuals on the
increase in ED visits between during 4 time intervals between 1996 and 2004.
Variables evaluated included insurance status, family income, and having a regular
source of care. The results demonstrated that increasing ED visits cannot be
attributed to lack of insurance or a primary caregiver. The proportion of ED visits by
uninsured decreased slightly from 15.5% to 14.5% during the study period. The
proportion of individuals who had a primary care physician actually increased from
52.4% to 59%; in contrast those without a primary care physician remained stable at
9.7% to 9.6% during the study. Use by individuals whose family income was >400%
of the poverty level also increased from 21.9% to 29% during the study period. The
authors had no explanation for the apparent increase in use by non-poor individual
with a regular source of care (E. J. Weber, et al., 2008). A common perception
related to the belief that increasing ED utilization is driven by individuals who are

unable to pay is the opinion that immigrants (legal or illegal) have caused rapidly
rising ED utilization. A unique study compared health care utilization (including ED
use) of adults on Medicaid with limited English proficiency with those who were
English proficient. During the two year study period utilization by 567 individuals
with limited English proficiency was compared with 1,162 individuals who were
English proficient. The results demonstrated that subjects with limited English
proficiency were enrolled longer and more continuously in Medicaid, were 94% more
likely to utilize primary care resources, and were 78% less likely to seek care in the
emergency department (Graham, Jacobs, Kwan-Gett, & Cover, 2008).
A study using the Health Resources and Services Administration's county-
based data evaluated the relationship between the proportion of local physicians
providing primary care (as a proxy for primary care access) with a variety of
measures of healthcare utilization including ED visits. The proportion of primary
care physicians within counties ranged between 20% and 54%, with a mean of 34%.
The study demonstrated a decrease of utilization, and estimated that a 1% increase in
primary care physicians resulted in an average reduction of 503 hospital admissions
and 2,968 emergency departments within the counties studied (Kravet, et al., 2008).
Other studies have also evaluated the impact of primary care physician access on
individuals decision to seek care in the emergency department. A study of Medicaid
patients attempted to determine whether a relationship existed between the office
hours of the patients primary care physicians and their tendency to seek care in the

ED. The study population was 57,850 HMO patients who received care 353 primary
care physicians; physician practice data were collected through a survey of the
physicians themselves, and matched to patients ED records based on the PCP
assigned to them through the HMO. Patient visits were coded as potentially
avoidable and probably unavoidable based on a previously validated scoring
method that identified cases that could have been adequately managed using a prompt
appointment with the PCP. The study population had 32,156 ED visits; 5,043
(15.6%) were classified as potentially avoidable. Patients from physician practices
that had 12 or more evening hours/week utilized the ED 20% less than other patients.
Although the association was not statistically strong, potentially avoidable ED visits
were less frequent for patients from physician practices with weekday evening and/or
weekend hours. The authors calculated that overall ED use could be reduced by 13%
if all physician practices offered 12 or more evening/weekend hours (Lowe, et al.,
The Emergency Medicine Patients Access to Healthcare (EMPATH) study is
a cross-sectional, observational study in 30 hospitals to identify why patients seek
care in the emergency department. The study involved the collection of demographic,
clinical, and insurance data, as well as a structured interview of all consenting
patients seeking care at all facilities during a single 24 hour period. The study yielded
2,011 eligible patients out of 2,874 total ED visits. Patient interviews were completed
on 1,579 (78.5%) subjects. Average age was 45.9 years; 44.6% were male and 55.4%

female. Subjects were predominantly white (58.3%); 28.3% were black, 7%
Hispanic, and 6% other. A total of 81.3% had insurance of some kind. Using a
frequency analysis of the interview data, the study revealed five factors that were
cited most frequently as the reason for seeking care in the ED. Medical necessity
(articulated by statements such as "this is a medical emergency, the ED is the right
place for this problem, too worried about my problem or too sick to go
elsewhere) was the most frequently cited (by 95% of participants). Additional
rationale included convenience (86.5%), ED preference (better medical care here,
get my regular care at this hospital) (88.7%), affordability (25.2%), and limitations
of insurance (14.9%) (Ragin, et ah, 2005).
The literature reviewed above has been based predominantly on data collected
on the US healthcare system. Although their results must be carefully considered to
ascertain their cultural relevance, a number of useful studies have been published on
data from other countries. A study from England on patients with primary care
problems who presented to the Accident and Emergency (A & E) department
offers some perspective. The sample population was 892 patients presenting to an A
& E department who were triaged as having a non-urgent problem. Eight hundred
fifty-five (855) agreed to be interviewed (96% response rate). The majority of
patients (58.9%) stated that they had a general practitioner (GP) nearby; of this group,
66.7% presented at the A & E department during hours when their GPs office was
open. The patients reasons for choosing the A & E department were revealing. The

most common reason for not going to the GP was a perception that the problem was
too urgent (24.7%). Of the patients with traumatic problems, 31.6% chose the A & E
department because they thought it would be inappropriate to go to the GP with a
traumatic problem. Dissatisfaction with the GP was a recurrent theme, as 14.9% had
been to the GP and were dissatisfied with their care, and another 7.4% had been so
dissatisfied with the GP in the past that they didn't even consider them this time.
Twenty percent (20%) believed that their GP would be unavailable, and another
14.6% had tried the GP and felt that the wait would be too long. Finally, 7.1% were
referred to the A & E department by their GP (J. Green, Dale, J, 1992). A second
study from England sought to evaluate the motives of patient self-referrals to the
Accident and Emergency department. Based on postal surveys from 344 patients
who self-referred, the authors found that the most common reasons for self-referral
were a perceived need for the diagnostic facilities available at the A & E, followed by
a belief that their problem could be best handled by the hospital physician. Subjects
also voiced dissatisfaction with their general practitioners as an additional reason for
self-referring to the A & E (Moll van Charante, ter Riet, & Bindels, 2008).
A small Swedish study used in-depth qualitative interviews to understand the
perspective of 10 frequent ED attendees (6-17 ED visits during the previous 12
months). The authors noted that all of the study subjects sought care in the ED
because of fear that their symptoms were life-threateningdespite the fact that many
were viewed by the staff as harmless. Associated with the threat of the symptoms

was a fear of loss of autonomy and even of death. The other common theme
described by study subjects was a life of stressful circumstances that included a broad
variety of social, psychological, family, alcohol-related, and/or health problems. The
authors proposed that this social context increased the degree of vulnerability that
subjects experienced with even minor symptoms (M. Olsson & Hansagi, 2001).
Another qualitative study from Israel used a 30 item questionnaire to explore
the characteristics of 74 individuals who self-referred to the ER, and to compare the
perceptions of the subjects with the nurses who treated them. The questionnaire
collected information on the subjects' demographic variables, non-clinical reasons for
seeking care in the ED, decision-making strategies, and the perceived urgency of their
symptoms. Orthopedic and minor complaints comprised 65.3% of the clinical
problems, followed by a variety of minor medical complaints. Thirty percent (30%)
of the respondents reported experiencing an unusual amount of stress and anxiety.
Over 60% reported seeking care in the ER because they believed the treatment there
was better than at other sites. The majority of subjects (77%) believed their condition
to be urgent or most urgent; the nurses caring for the subjects evaluated 78% of the
subjects conditions to be non-urgent (Rassin, Nasie, Bechor, Weiss, & Silner, 2006).
Several Canadian studies also provide insight into patterns of ED use. Afilalo and
colleagues performed subject interviews on participants identified through a
secondary analysis of a cross-sectional study of five ERs in Quebec. Based on the
use of five point acuity scale subjects were divided into urgent and non-urgent

groups; a study aim was to compare these two groups. The majority of both groups
reporting having a primary care provider (69.8% non-urgent, 75.2% urgent, no
significant difference). Non-urgent subjects were slightly younger than urgent (43.3
vs. 48.7) and were more likely to visit the ED on the weekend or after hours. The
non-urgent subjects identified the following five reasons for seeking care in the ED:
inability to access primary caregiver (32%), perceived medical need (22%), referral
by PCP (20%), familiarity (11%), and trust (7%). The authors noted that the absence
of any predominant rationale for choosing the ED might explain why various
strategies to divert patients from the ED to other sites have been so unsuccessful (J.
Afilalo, et al., 2004). Survey data from a random sample of patients from two E.D.'s
in Edmonton evaluated subjects attempts to access care elsewhere prior to going to
the ED. From a sample of 1,389 randomly selected patients 905 (65%) agreed to be
surveyed. Sixty-one percent (61%) reported seeking care elsewhere prior to coming
to the ED. Approximately 17% of the individuals who reporting seeking care
elsewhere stated that they felt the ED was the best option for their needs (Han,
Ospina, Blitz, Strome, & Rowe, 2007).
Alternatives to the Emergency Department
Patients with non-emergent acute complaints seek care in locations other than
the emergency department, including their primary care provider and walk-in clinics
or urgent care clinics. Because of the variety of reasons individuals seek care from
their primary care provider (preventative care, examinations, vaccinations, acute

symptoms) there is very little data on the patterns of primary care provider use for
acute complaints. According to the National Ambulatory Medical Care Survey 2005
symptom complaints comprise 47.5% of all office visits; of the 47.5% respiratory
symptoms are the most common non-traumatic complaint contributing 9.9% (Cherry,
Data regarding utilization of walk-in or urgent care clinics comes from a
variety of sources. An early study evaluated the demographic attributes, attitudes,
and perceived barriers to primary care reported by individuals using an urgent care
facility within an HMO. The cross sectional survey of 421 adults who sought care in
the urgent facility demonstrated that urgent care users were younger than the primary
care patients (40 vs. 46) but were otherwise similar. Almost 90% had a primary care
provider, and an overwhelming majority made very positive statements about primary
care and their primary care provider: I think it is important to see the same doctor
over time (88%); 1 am satisfied with the care I get from my regular doctor (86%);
My doctor is available to answer my questions whenever I need him or her (68%).
The study found that only 15% agreed with the following statement: I prefer urgent
cares doctor to my assigned doctor for this problem. Rationale for seeking care in
the urgent care department included needing to be seen immediately (64%), because
the primary care offices were closed (47%), work or childcare constraints (27%), and
unable to get an appointment with primary care physician (25%). Forty seven percent

(47%) stated that they would prefer to see their primary care physician w ithin one to
two days rather than attending the urgent care clinic (Plauth & Pearson, 1998).
Because of their emphasis on the use of general practitioners and clinics rather
than the Accident and Emergency Department, Britain's National Health Service has
done several related studies. A study was performed by Jackson et al. to determine
how people choose and use the NHS Walk-in Centres which were developed to
reduce the patient load in the Accident and Emergency Department as well as in
primary care facilities (C. J. Jackson, Dixon-Woods, Hsu, & Kurinczuk, 2005). All
individuals attending the Walk-in Centre during the five week study period were
invited to participate in the study. Forty five (40%) agreed to participate. Purposive
sampling was utilized within the 45 volunteers to assure adequate representation of
clinic hours and subject gender; 23 individuals were interviewed. Qualitative analysis
revealed three key themes related to patterns of service use: seeking help, resources,
and access. The authors described two types of service use-: those who knew
what was wrong with them and had a clear idea of the required treatment (referred to
as execution and implementation" in the study) and those who were seeking
professional advice (referred to as professional advice). Individuals whose style for
seeking care was execution and implementation were described by the authors as
individuals who ...had decided or knew what was wrong with them and had a clear
idea of the type of treatment they required. Their main purpose in help-seeking was
to seek assistance in executing a plan of action on which they had already determined,

either by themselves or on the advice of a health professional. The authors went on
to note that These patients chose the service that they believed would give them
what they required." In contrast, individuals whose style for seeking care was
professional advice were uncertain about the nature of the problem, and particularly
whether or not it was serious. They sought professional advice, along with treatment
if necessary, and often wanted to resolve uncertainty about an anxiety-provoking
Prior studies have suggested that the decision to seek care in the walk-in
(urgent care) setting is primarily based on a desire for convenience (Salisbury, 2002).
However, in their discussion, Jackson et al. suggested that convenience was not the
primary decision factor for their subjects. They proposed that an additional factor
was what they referred to as social positioninga combination of concern
regarding the impact of their selection of site on the demands of the facility/
practitioners, and fear of being seen as an inappropriate user if they sought care in
the wrong location. They noted that in many cases their subjects saw their general
practitioner as the ideal source for care but, like the subjects in the current study,
chose to not seek care there. They proposed two different reasons that subjects chose
to not seek care with their preferred practitioner: scheduling delays and a desire to not
further overload their GP. The issue of scheduling delays has been described
previously. It is important to note that the significant differences in health care access

and delivery between the US and UK systems may limit the applicability of these
findings to current study.
Fear of being an inappropriate patient was mentioned by Jackson et al. (C. J.
Jackson, et al., 2005), and has been the subject of several other studies, especially as
it relates to use of the emergency department. In an early qualitative study of
deviant patients in the emergency department Roger Jeffery began by describing
good patients as individuals who: enable the physician to practice skills necessary
maintain his/her skills; allow the physician to practice his/her chosen specialty; and
have a challenging problem that requires intervention. All other cases were what
Jefferys subjects (physicians practicing in the Accident and Emergency department)
referred to as rubbish. He went on to describe how staff punished patients who
were rubbish through delays in treatment, being placed in rooms where they could
be ignored for prolonged periods of time, or receiving verbal hostility (Jeffery, 1979).
Other studies by Sanders also described how emergency department staff are less
motivated to care for inappropriate ED patients (Sanders, 2000) or displayed
negative feelings, irritation, and impatience with patients who presented to the ED
with what they considered primary care problems (Crouch & Dale, 1994). Jackson
et al.s subjects described similar concerns regarding the stigma associated with
wasting the PCP or emergency departments time. They also described their
relationship with their GP as a valuable but scarce asset; the researchers postulated
that the Walk-in Centre offered an alternative to the GPs office that was practical but

did not disrupt the valuable relationship between the subject and his/her GP (C. J.
Jackson, et al., 2005).
A study from Switzerland compared patients who attended a university-based
walk-in clinic with those who sought care from their primary care provider.
Questionnaires were completed by 329 subjects, and qualitative interviews were
performed on 26 individuals randomly selected from the questionnaire sample.
Reasons for selecting a facility were quite different between the two groups;
individuals seeking care in the private practice relied on recommendations from
members of their social network, while subjects from the outpatient clinics reported
being referred by a health professional or for location or convenience reasons. They
had different reasons for remaining with their site as well. Outpatient clinic subjects
chose to stay because they appreciated the equipment the clinic had, the technical
skills of the staff, or the convenience. In contrast, the private practice subjects valued
the personal care, feeling known" by the staff, and friendship (Perron, Secretan,
Vannotti, Pecoud, & Favrat, 2003).
A variety of factors appear to be related to the decision about where to seek
care. Different perceptions of urgency, quality of care, access, convenience, facility
capability, and care provider attributes combine to guide individuals to the emergency
department, urgent care facility, or primary care provider. Appendix 1 summarizes
the pertinent research on the subject.

The studies cited above describe individuals seeking care in the E.D. with the
following characteristics:
Choices regarding where to receive care.
A belief that the E.D. is a more competent site to receive care, either because
the complaint is perceived as "too serious for primary caregivers, or because
the patient is dissatisfied with prior primary care experiences.
A belief that the E.D. is the most convenient location to receive care.
In most cases, some form of health care insurance.
A willingness to endure some delays and discomforts in order to receive the
desired care in the E.D.
A need to receive clear and specific instructions concerning post-E.D. care,
medications, limitations, and follow-up appointments.
Are there explanatory models available to describe why, how, and where
individuals with these characteristics seek care? When deconstructed, the seemingly
simple act of seeking care for a medical condition is composed of several independent
but important factors. First, the individual must interpret a given set of symptoms as
an experience worthy of seeking care; this interpretation is deeply rooted in
individual, cultural, and social definitions, and will be explored below. Second, the
individual must recognize and have access to sources of care; more directly, the act of
seeking care must not be restricted by any barriers. In this context, sources of care

are broadly defined to include popular, folk, and professional sectors as described
below. Finally, the individual must decide where and how to seek care. The
theoretical foundations for each of these factors are reviewed below.
The Meaning and Interpretation of Symptoms
What causes an individual to seek care for a symptom? The answer to this
simple question is based on a complex cultural nomenclature that describes the
relationship between the daily discomforts of life and how they impact activities of
daily living. "Illness" writes Allen Young in Anthropologies of Illness and of
Sickness refers to a persons perceptions and experiences of certain socially
disvalued states, including, but not limited to, disease" (Young 1982). More simply,
Cassell uses the word illness" to represent what the patient feels when he goes to
the doctor (Cassell, 1976). Is illness, as defined by Cassell, simply the presence of
ANY abnormal symptom? Probably not. Multiple studies have demonstrated that the
vast majority of individuals experience abnormal changes in body function weekly,
but very few actually seek care (Apple, 1960; Dunnell, 1972; Stoeckle, 1963). In a
1961 summary of studies from the United States and United Kingdom, White et al.
reported that during any given month approximately 75% of the population reports
having symptoms (being sick or injured) (White, 1961); in an update of this study
using 1996 data, Green et al. increased the estimate of individuals experiencing
symptoms each month to 80% (L. A. Green, Fryer, Yawn, Lanier, & Dovey, 2001).
The perception of these symptoms as "illness is based on a combination of the

perceptions of self and others, all within the cultural norms of the group. Defining
"illness typically involves one or more of the following: perceptions of change in
body appearance; change in body function; unusual body emissions or secretions;
changes in the function of the arms or legs; alterations in sensory perception (hearing,
vision, touch); physical symptoms (headache, shortness of breath); emotional states
(such as anxiety); behavioral changes (work behavior, relationships to others)
(Helman, 1981).
Interpretation of these factors as either illness or "normal is remarkably
variable among cultures. Medical anthropologists have documented a broad variety
of interpretations of "normal symptoms between various racial groups within the
United States, as well as between various countries and cultures worldwide (Helman,
1981). Although it is beyond the context of this review, the unique culture of the
North American professional medical community has re-interpreted the "normalcy
of a broad range of symptoms such as menopause, pregnancy-related symptoms, and
age-related changes as being abnormal and requiring treatment; this is a markedly
different approach to these symptoms when compared to elsewhere in the world
(Lock, 2000). Overall, the convergence of personal perceptions and cultural norms
creates the definition of "illness when there is agreement between the individuals
perception of impairment and that of the individuals family and social context
(Helman, 1981).

Once the individual is defined as 'ill several actions typically follow. First,
most cultures define a sick role which encompasses behaviors that the individual
adopts in response to specific culturally defined expectations; these behaviors are
heavily influenced by the expectations of care providers (McElroy, 2000). In most
cultures the individual is not considered to be responsible for his/her illness, and is
allowed to withdraw from normal adult responsibilities as long as attempts are
made to regain health (Parsons, 1948). Within the sick role individuals may
receive care from one of several culturally defined sources of health care (Fox, 1968).
Kleinman has identified three sources for health care in any society: the popular, folk,
and professional sectors (A. Kleinman, 1980); these sources will be further described
Symptoms have meaning to individuals, societies, and each of the sectors of
healthcare described above. Kleinman has suggested a framework known as the
Explanatory Model to explain the interpretation of symptoms by all healthcare
delivery sectors (Kleinman 1980). The Explanatory Model (EM) offers explication
for illness, guidance for treatment, and a framework in which to organize the social
meaning of the illness itself. The EM is organized around key questions that
individuals and societies must answer about a symptom including (A. Kleinman,
Seeman, D, 2000):
What is the problem? Is it an illness?
How does it affect the body-self?

What can be expected to happen next?
What will be the long-term outcome?
Will it get better or worse?
What is to be most feared about this condition?
What treatment is most appropriate?
What is most to be feared about the treatment?
The answers to these questions, and the meaning of the answers, are in part
driven by the cultural and social milieu of each of the healthcare sectors. In the
professional and folk sectors, these questions are largely driven by the history and
physical exam blended with the patients past history and risk factors; they aid in the
search of an identifiable diagnosis as defined by the knowledge and science of
each sector. These sectors seek to answer the last two questions about treatment
plans based on acceptable practice (again within their individual traditions) and
especially in the case of the professional sectorprobabilities. For the popular
sector, the answers to these questions are largely driven by the cultural and social
meaning (if any) of the symptoms combined with any previous individual or social
experience with these symptoms in the past; this often involves intersection with the
folk and/or professional sectors.
In summary, symptom interpretation is largely determined by cultural norms
as interpreted by family members, and provides the foundation for all decisions
related to healthcare seeking behavior.

Knowledge of and Access to Sources of Healthcare
Once the individual has determined that a symptom warrants care, a site for
that care must be identified. This choice is influenced by both knowledge of and
access to potential care sites, as well as the opinions of the individual's social and
cultural contacts (Stoeckle, 1963). Knowledge of care sites, like the meaning of
symptoms, is largely based in social interactions with family and community. Rubel
and Garro refer to the "understanding and information people have from family,
friends, and neighbors as to the nature of a health problem, its cause, and its
implications" as the health culture (Rubel & Garro, 1992). Depending upon the
health culture, a given symptom may be dealt with by waiting for it to "run its
course", applying self-care, receiving care from a family member, or by consulting
with a care source outside of the family. Thus, symptoms are interpreted by the
individual and family members, and decisions are made whether to seek outside
counsel or treatment. A number of studies have suggested that most symptoms are
managed within the household without "professional" assistance (L. A. Green, et al.,
2001; A. Kleinman, Eisenberg, L, Good, B, 1978; McElroy, 2000; White, 1961).
Kleinman has identified three sources for health care in any society: the popular, folk,
and professional sectors (A. Kleinman, 1980). Popular health care encompasses all
health care considerations that do not involve either socially-defined folk healers or
medical professionals. In the vast majority of cases, popular health care is associated
with family resources (Helman, 1981), and it is within the family that the individuals

perception of illness is confirmed (or denied), and that initial treatments are started.
Kleinman et al. estimate that 70%-90% of all care occurs at the self or family
treatment level, regardless of the culture (A. Kleinman, Eisenberg, L, Good, B, 1978).
In a more recent study, Green et al. estimate that 40% of individuals experiencing
symptoms consider seeking care, but that only 35% actually do, resulting in 65% of
symptomatic individuals getting care at the self or family level (L. A. Green, et al.,
2001) Based, in part, on the assessment made at the popular level, additional care
may be sought at "higher" levels.
The folk sector of health care is made up of individuals who are identified
within the culture as having special knowledge of health-related matters, but who are
not part of the professional health sector, as defined by education and/or licensure.
The basis of the folk healers special status may be either spiritual or secular.
Herbalists, shamans, midwives (non-nurse), and spiritual healers are all examples of
folk health practitioners. While most cultures have representatives of this sector,
there is broad variation in the nature and role of these individuals within society. As
compared to members of the professional health sector, folk healers are more likely to
be part of the community, and to share in the health perceptions and language of that
community. In most cases, the folk and professional sectors view one another with
disdain and distrust (Helman, 1981).
Finally, the professional health sector is made up of Western allopathic
physicians (including medical doctors, doctors of osteopathy, and chiropractors), as

well as allied health professionals such as nurses, respiratory therapists etc. Despite
the Western obsession with the professional health sector, it is believed that
worldwide only a small fraction of all healthcare is delivered by this sector; the
majority of healthcare is provided by the combination of popular and folk approaches
(Helman 1982).
Access to folk and professional healthcare services is not universal. The
ability of an individual or group to obtain the timely use of personal health services
to achieve the best possible health outcomes is known as access (Institute of
Medicine, 1993), p 4). A population's access to healthcare may be evaluated by
looking at a variety of direct and indirect variables including frequency of visits to
health practitioners, who provided the care, where the care was delivered, and the
purpose and nature of the visit. Access to care might also be indirectly evaluated by
looking at outcome measures such as mortality, death rates, disease incidence, and
hospitalization rates (Institute of Medicine, 1993), p. 36). When access appears to be
reduced in a given population, queries are appropriate to determine if there are
barriers that prevent individuals or groups from having access to necessary healthcare
Barriers to access to the professional sector of healthcare may be structural,
financial, or personal/cultural (Institute of Medicine, 1993)p. 39). Structural barriers
include any factors that reduce the availability of healthcare resources for a given
population. Structural barriers may include insufficient numbers of practitioners,

healthcare delivery sites that are difficult or impossible for the population to access,
or organizational factors within the healthcare delivery system that make it more
difficult to utilize. In many cases, these structural barriers have their origin in the
methods that healthcare is financed. Financial barriers impact consumers by either
reducing the affordability of care or making it financially unattractive for healthcare
providers to offer services in a particular geographic area or population segment. The
rapid escalation in healthcare cost and the influence of poverty are both powerful
contributors to the decreasing affordability of American healthcare. Financial
pressures on providers include an unwillingness or inability to provide
uncompensated care for impoverished or uninsured individuals and/or reimbursement
schedules that make it unattractive to provide services to a given population (such as
Medicaid or Medicare recipients). Personal/Cultural barriers are factors that may
lead a given population or subgroup to under-utilize healthcare services, or to fail to
follow the recommendations provided when they do receive care. Cultural factors
that may impede access to healthcare include language, beliefs about the value of
"professional as compared to folk healthcare, or misperceptions about what actually
occurs within hospitals or other health settings. The individual's health culture, as
defined earlier, has a profound impact on decisions related to illness management
including seeking care and complying with recommendations of healthcare
professionals (McElroy, 2000; Rubel & Garro, 1992). For individuals and
organizations dedicated to the elimination of barriers to equal access for healthcare

services (primarily related to access to professional sector healthcare), it has proven
difficult to separate the relative influence of these three barriers to healthcare access,
particularly in populations that are most disadvantaged because of poverty, minority
status, and social displacement.
Like symptom interpretation, knowledge of and attitudes related to various
sources of healthcare are largely determined by family interpretation of cultural
norms within the health culture. Once these various sources are recognized, the
individual's ability to utilize them may be influenced by a variety of factors that may
reduce access.
Selecting the Healthcare Site
For individuals with culturally significant symptoms and adequate access to
healthcare resources a decision must be made regarding where to seek care. In a
perfect world, all decisionsincluding those regarding health carewould be made
in a rational and logical fashion by individuals who know their consistent preferences,
are fully informed of the alternatives available, have knowledge of the strengths,
weaknesses, and consequences for selecting each alternative, and combine the
information that is available based on reasonable probability theory (Dawes, 1988;
Fischoff, 1975). The next sections provide an overview of various decision theories,
followed by theories related specifically to healthcare decisions.

Decision Theoiy
Theories describing human decision making may be organized according to
assumptions related to the limitations of the human mind. Early philosophers,
including Pierre Simon Laplace and John Locke, believed that the human mind was
without limit; they believed that all human decisions could be explained through
probability theory, Bayesian models, and the analysis of available data to achieve
maximum expected utility. These theories of unbounded rationality" have recently
been countered by those who suggest that there are limits to the human mind
(bounded rationality) (Todd & Gigerenzer, 2000). A review of these two
fundamental approaches offers a useful context for discussions concerning health
decision making.
Unbounded Rationality
While early considerations of the human mind suggested that there were no
limits to its capacity, a more practical approach suggests that even if
computational power were infinite available data and computational time are in
fact finite. Lor this reason, contemporary unbounded rationality theorists suggest that
daily decision-making is an example of optimization under constraints. Proponents
of optimization under constraints suggest that in the real world search must be
restricted due to limitations of time, information, attention or money (to pay others to
make the decision). The key to optimization under constraints is what is referred to
as a stopping rule, a way to determine when to stop searching for information

because the "cost" of additional search exceeds the potential benefit (Sargent, 1993;
Todd & Gigerenzer, 2000). For example, a "stopping rule" for an individual seeking
a healthcare site might be to stop" seeking as soon as a site is found that is available
at a time convenient to the individual. Alternatively, the "stopping rule" might be to
stop seeking as soon as a site is found that has received high scores from a local
published report on healthcare quality. A third "stopping rule" could be to stop
seeking as soon as a site is found that is covered by an insurance plan. Obviously, in
this example, which site is selected will be entirely dependent upon the stopping rule.
Opponents of the optimization under constraints approach suggest that selection of
the stopping rule itself may require an unrealistic amount of knowledge and
computation (Vriend, 1996).
For supporters of unbounded rationality there is another issue to confront: the
apparent illogic of many real decisions. If the assumption is made that all human
decisions are made from the perspective of maximum utility and benefit to the
individual, it may be difficult to explain why humans appear to make so many bad
decisions. Even if optimization under constraint is considered, many human
decisions (examples might include the decision to smoke cigarettes, not wear
seatbelts, or spend more money than is earned) appear irrational (Todd & Gigerenzer,
2000). A substantial body of data exists that suggest that actual decisions are often
not consistent with a rational model. In a classic study Meehl (Meehl, 1954)
compared actual clinical judgments with accepted formulas; he found that the

formulas were consistently more accurate. The study also compared clinician's
estimate of their own decision-making performance with their actual accuracy and
found that they were supremely confident of their decision-making skills despite very
modest actual success. So, if decision-making were based on rational methods, they
weren't very accurate. In another pivotal study, Edwards (Edwards, 1968)
demonstrated that intuitive assessments of likelihood (a key aspect of rational
decision-making) were inconsistent with the norm predicted by rational theory A
variety of other researchers have also discovered that there are consistent errors in
actual human decision-making that are very inconsistent with rational" thinking
(Bazerman, 1986; Dawes, 1988; Hogarth, 1987; Kahneman, 1982a, 1982b). One
author makes the following statement about the application of expected utility and
rational choice explanations for human decision making:
We propose that expected utility and rational choice models are
most likely to do a good job of predicting decision maker's choices
under three conditions: (a) when the choice environment is less, rather
than more, social; (b) when the situation or context makes the
economic structure of a decision particularly salient; and (c) when the
context calls for a deliberate, calculating approach to decision making
(J. M. Weber, Kopelman, & Messick, 2004).
This statement brings the usefulness of the expected utility and rational choice models
into question; healthcare decisions are extremely social due to the influence of the

health culture (McElroy, 2000; Rubel & Garro, 1992), are not typically made based
on economic constraints (Marquis, et ah, 2006), and may require decision processes
that lack time and information required for a deliberate, calculating approach
(Hibbard, 1997).
Bounded Rationality
What could explain the frequency of inaccurate decisions within a rational
framework? Obvious causes could include inadequate information, confusion about
the qualities of a successful decision, or even apathy on behalf of the individual
making the decision. However, an alternative explanation that has garnered
significant support during the past 30-40 years is that the human mind has inherent
limitations for holding, searching, and processing information. An early author
describing these limitations referred to human decision-making as bounded
rationality suggesting that individual rationality is "bounded by perceptions of a
situation, point of view, and limited computational abilities (Simon, 1957).
Proponents of this view believe that humans intuitively simplify situations and
narrow their viewpoints to align decision-making with their cognitive limitations. To
quote one author, People reason and choose rationally, but only within the
constraints imposed by their limited search and computational capacities (Simon,
1957). At its core, the bounded rationality approach is based on the assumption
that there are limitations to human cognition, an assumption that could prove

important to individuals making decisions in an increasingly complex health care
Within this assumption there are two basic frameworks to explain how the
human mind deals with the limitations of its own computational capacity and the
limitations of environmental information. The first is known as satisficing and the
second heuristics. Satisficing is a method for selecting between a group of
alternatives about which the individual has little information. The essence of
satisficing is identifying the characteristics of a satisfactory choice and then selecting
the first alternative that meets those characteristics (Simon, 1990). For example, an
individual with little information about automobiles might choose by purchasing the
first vehicle encountered that met their primary characteristic: a price of $ 15,000 or
less. Although satisficing provides a strategy for selecting that stays within the
limitations of time and environmental information, it may still require analysis or luck
to determine the characteristics of a satisfactory choice (Todd & Gigerenzer, 2000);
in the automobile example, a price that is too low will likely lead to an ultimately
unsatisfactory selection, while one that is too high will potentially provide less value
for too high a cost. In healthcare, consumers frequently have too little information to
make an appropriate decision about what a satisfactory outcome could be.

While satisficing offers a strategy for rapid decision-making in an
environment with limited information, potential uncertainty about the characteristics
of a satisfactory solution may limit its usefulness in explaining or predicting health
decisions. An alternative theoretical strategy for dealing with decision-making
limitations is the creation of shortcuts or rules of thumb" to improve the quality of
decisions (Carroll, 1990). These rules of thumb, or heuristics, offer another approach
to making decisions with limited computational power and time. In contrast to
satisficing, which selects the first alternative that meets the individual's established
conditions for satisfaction, heuristics are created when the individual combines prior
experience with simple rules of thumb. Heuristic theorists refer to the role of prior
experience in decision making as bias.
A significant body of research exists to describe the presence of bias in human
decision-making. In their book Judgment under Uncertainty: Heuristics and Biases
Kahneman, Slovic and Tversky proposed that, while task complexity and limited
processing capacity are issues for human decision-making, the creation of heuristics
rules of thumb" to guide thought processes represents an entirely different
type of process (Kahneman, 1982a). The central premise of their argument is that
judgment under uncertain conditions is based on a number of simplifying heuristics
rather than an extensive algorithmic process (Gilovich, 2002). Each heuristic is
created by the individual in association with a bias that serves as the basis of the rule

of thumb." Some theorists postulate that some of the best heuristics may be based on
a single rationale or reason such as whether any existing options have ever been
encountered before (Gigerenzer, 1999; Todd & Gigerenzer, 2000) and, if they have,
whether they meet other characteristics of success.
Heuristic theorists believe that, while individuals may develop specific
heuristics and biases to meet their unique needs, there may exist universal"
heuristics that can explain many common judgments (Gigerenzer, 1999; Kahneman,
1982a). While it is beyond the scope, and not consistent with the purpose, of this
study to chronicle all of these heuristics, several have particular pertinence to how
individuals make healthcare decisions: recognition, anchoring and adjustment,
availability, and representativeness .
Humans have a remarkable capacity to identify things that they have
encountered before, even if they cannot remember specific details (Gigerenzer, 1999).
This capacity is the foundation for the recognition" heuristic, which encompasses
two distinct products of experience that may assist in making a decision. The first is
recognition of the situation. The more familiar a decision maker is in a given
situation the more rapid and confident the decision making will be (J. M. Weber, et
al., 2004). While this is intuitive, it is also important because it guides the decision
maker to recall the similar situation, remember the solutions utilized previously, and
ask whether those solutions were satisfactory. A corollary heuristic known as take

the last" suggests that an individual will start the search for a solution with the
strategy that was perceived to be successful in the last encounter with a familiar (or
similar) situation. In many cases individuals will repeat this strategy until it fails
(Gigerenzer, 1999). The combination of "recognition" and "pick the last" heuristics
may become the foundation of a response habit that, over time, may not result in
maximum benefit to the individual (J. M. Weber, et al., 2004), such as when a patient
routinely requests (and/or a physician routinely prescribes) antibiotics for all episodes
of cough because they appeared to be effective in a prior similar episode.
The second aspect of the "recognition" heuristic is recognition of an option or choice.
In the situation when the decision maker is NOT familiar with a situation, a decision
may need to be made among several options. If the decision maker has no specific
knowledge of the options the "recognition heuristic suggests that the decision-maker
will select any option that is recognized within any context (Gigerenzer, 1999).
The recognition heuristic has broad applicability within healthcare. An
individual experiencing a symptom may consider all the possible causes of the
symptom, then do a literature search to determine the most likely cause and course of
action (an unbounded rational approach). More realistically, use of the recognition
heuristic would guide the individual to attempt to remember prior experiences with
the same symptom and, if remembered, to consider what was tried previously and
whether it worked. Likewise, for the decision maker who is attempting to select a
healthcare facility from a list, the recognition heuristic suggests that preference will

be shown to any that the individual recognizes even if the context surrounding the
recognition cannot be remembered.
Anchoring; and Adjustment
Kahneman, Slovic, and Tversky's Anchoring and adjustment heuristic
(Kahneman, 1982a) is closely related to "recognition." This heuristic suggests that
individuals who are confronted with an unknown decision anchor" their choice on
information that IS known, then adjust" it according to their best guess. So, an
elderly woman who has fallen will decide whether and how to seek medical
assistance based on what she knows about a friend of similar age who had a similar
experience. In her friends case, she chose to call an ambulance to be transported to
the emergency department; the emergency department found that she had only minor
injuries and she returned home the same day. However, despite the absence of
serious injury, her friend received several medical bills for thousands of dollars. In
this case, the friends experience (sort of a third-party recognition) is the anchor,
which will be adjusted by the individuals understanding of how her experience might
be different from that of her friend. Of course, the opportunities for error within this
heuristic are multiple: the friends experience might have been unusual, or their
conditions might be very dissimilar. Nevertheless, the anchoring and adjustment
heuristic provides a starting point for making a rapid decision that can be based on
some element of prior experience.

Another of Slovic, Tversky & Kahneman's general purpose heuristics is
availability. This heuristic provides individuals with an estimate of the likelihood of
an event "by the ease with which instances or associations come to mind" (Schwarz
N, 2002). A middle aged man who has chest pain might immediately assume his
problem is a heart attack because he has recent memory of several friends who have
experienced cardiac problems and cannot think of any other associations to "middle
aged man with chest pain. Likewise, an individual with an elderly father who has
developed memory problems might immediately assume a diagnosis of Alzheimer's
based on the plethora of television and print media pieces that have been published in
the recent past. Like anchoring and adjustment, the availability heuristic provides a
biased understanding of a phenomenon that may not be accurate.
The final general purpose heuristic described by Tversky, Slovic and
Kahneman is representativeness. This heuristic has application in virtually all other
heuristics because it involves an individuals assessment of their similarity to the
other individuals (or situations) that may be considered using anchoring, adjustment,
or availability. So, for instance, a man experiencing abdominal pain may anchor the
experience with a similar symptom his wife had. The attributes of the pain are very
compatible with little or no adjustment. However, he rapidly concludes that her
experience is not representative of his because her abdominal pain was the result of

pregnancy. Representativeness may involve similarity in attributes between the
individual and others, or may be related to causal factors or even belief sy stems. The
essence of the heuristic is an estimation of how well the individual and the
comparison group or experience correspond to one another (Tversky, 2002).
The Role of Identity and Social Norms in Decision Making
The importance of the "health culture" has been previously discussed.
Another limitation of theories of unbounded rationality is the absence of any
allowance for the role of social or emotional factors on "rational" decisions (J. M.
Weber, et al., 2004). Social factors could include the individual's self-identity,
information and expectations from family or significant friends, cultural norms, the
availability of financial resources, guidelines from insurers, and prior personal
experience. Emotional factors that might impact decisions could include fear, stress,
or denial. Theories of unbounded rationality propose several ways that social or
emotional factors may impact decision making. First, they may limit the pool of
options that the individual considers to be relevant to the situation, thus simplifying
the decision (J. M. Weber, et ah, 2004). For instance, an individual with severely
limited financial resources might only consider seeking care in health care facilities
that do not require payment upon receipt of services. Likewise, a young adult whose
family has always sought care from a local curandera may not even consider seeking
care in an urgent care or hospital facility due to the influence of family or social

Social or emotional factors may also impact decision making processes by
creating unique "rules of thumb." In fact, social norms themselves have been defined
as "understood rules for accepted and expected behavior" (J. M. Weber, et ah, 2004)
p. 284. Social norms are learned and imitated, and may establish acceptable
responses to specific situations (Gigerenzer, 1999). For example, social norms may
define when (if ever) and individual seeks care in the hospital, who assists in
childbirth, and which folk remedies should be utilized for common symptoms. In a
variety of settings social norms have been found to be extremely predictive of actual
decisions and behavior (Kallgren, 2000). Strong emotions may also impact decision
making by either shortening the number of options considered for a solution (denial
of the possibility of breast cancer causes a young woman with a lump to NOT include
going to the doctor among a list of potential strategies) or to create specific heuristics
(fear that chest pain might signify a heart attack results in a rule" that all chest pain
should be cared for in the emergency department (Gigerenzer, 1999)).
As noted previously, the decision whether and where to seek care for a
medical condition is a classic example of judgment under uncertainty. The individual
typically has significant deficiencies in diagnostic data, potential causes of symptoms,
implications associated with those causes, or potential sites where care could be
sought. While the rational model for decision-making is attractive, it appears
unrealistic to assume that consumers would have the requisite informational
framework necessary to make such a decision. However, virtually all individuals

have an experiential database on which to draw from for better or for worseto
create "rules of thumb.' The creation of heuristics, influenced by previous
experience, the healthcare culture, social norms, and emotions, provides a mechanism
for individuals to make rapid decisions in an environment lacking in information and
time. The next section will review behavioral theories specifically related to health.
Health Behavior Theory
Complex health care decisions offer an excellent example of how heuristics
may assist individuals to overcome many of the limitations of human decision-
making. Consumers begin with perceptions of their condition, which may be
inaccurate because of the lack of diagnostic tools or knowledge. These perceptions
become the foundation for deciding whether or where to seek care. The decision will
necessarily not be informed by a full list of potential diagnoses, nor of the clinical
implications of each diagnosis. In addition, many consumers have limited knowledge
of possible sites to seek care. Severely limited by lack of information, limited
perspectives, and implications of each choice, the consumer must simplify the choice
by using any previous experience and rules of thumb which, when combined, create
a personalized bias.
Within the health behavior literature, the Theory of Reasoned Action/Planned
Behavior holds similar assumptions. Originally developed by Fishbein (Fishbein,
1967), the theory is an attempt to discover the relationships between attitudes and
health behavior. The theory is based on the assumption that individuals take rational

steps in response to situations, and proposes that, when confronted by a specific
situation, the individual will process information and act in an apparently reasonable
fashion (Montano, 1997). In essence, the theory suggests that the intention of any
behavior is based on the attitude toward the behavior and the subjective norm
concerning that behavior. In more detail, the attitude toward the behavior is
determined by behavioral beliefs and evaluations (rational) about the outcomes
expected related to the behavior in question. The subjective norm is a function of the
individual's normative beliefs (which may be influenced by significant others) and
the motivation to comply with the norm. These influencing factors create the
behavioral intention that, unless confounded by outside factors, creates the behavior
in question (I. F. Ajzen, M, 1980). Application of this theory has some limitations.
First and foremost, it is primarily designed to predict behavioral intention, rather than
actual behavior. In addition, the theory assumes that the individual has complete
control of all behavioral choices (Montano, 1997), an assumption that is unlikely in
complex health decisions such as where to seek care. The Theory of Planned
Behavior is an extension of the Theory of Reasoned Action that incorporates the
ability of the individual to carry out the behavioral intention (I. Ajzen, 1991).

In a preliminary study of the factors used in deciding where to receive care for
non-emergent acute medical conditions, the author performed two qualitative
interviews of patients who had sought care at different times in a variety of settings
including the emergency department, urgent care clinic, and physician office (Bourn,
2002). One of the interviews was transcribed and the factors associated with
decision-making were coded. Several diverse themes emerged in the coding of the
Autonomy. The interview revealed that autonomy played a role in decision-making.
There was a loss of autonomy when going to a physician office because of the need to
make an appointment. A different loss of autonomy occurred in the E.D., where the
interviewee described a loss of control over the treatment plan as compared with the
physician office.
Issues of Competence & Quality. The interview revealed a broad spectrum of closely
held beliefs about the relative competence of each type of facility. For instance, there
was a perception that the E.D. provided more rapid diagnosis and treatment,
especially for life-threats and traumatic injuries. The physician office, on the other
hand, offered a more personal approach to care because the physician has a
relationship with the patient, as well as access to medical history and prior medical
experiences. While the E.D. was viewed as competent in virtually all conditions, the

physician office was restricted to non-traumatic, minor (as perceived by the patient)
Intervention. Each setting was felt to provide adequate intervention for the conditions
that were within its scope (as defined above). However, the physician office was
viewed as offering slower interventions that might not be effective, ultimately
requiring an E.D. visit anyway. The E.D. was perceived as offering a more
''professional" approach, immediate "cures", and an ability to do more comprehensive
evaluation because of the availability of "heavy equipment" (such as CT scan and/or
MRI) and on-call specialists.
Convenience. Again, each site offered its own conveniences. The physician office
and urgent care center were more conveniently located, and care in the physician
office faster (less waiting) once an appointment was made. However, the need to
make an appointment, usually 2-3 days after calling, was viewed as a major
inconvenience. The emergency department offered the convenience of being open
24/7. However, the need to wait often-prolonged periods of time to be seen was
viewed as a major inconvenience.
Cost. The individuals interviewed described adequate financial resources to pay for
care as necessary. One was essentially self-insured, while the other belonged to an
HMO. Both were willing to pay whatever was necessary for problems that they
perceived as being severe. Both were willing to pay for initial evaluations and
intervention in order to get a referral to the best physician. Within this context, it

was also clear that they wanted the cheapest alternative that would effectively resolve
the problem. One specifically noted that if he "self-diagnosed" the problem, he
wanted the least expensive care necessary to resolve the problem. Conversely, if he
couldnt self-diagnose, he was willing to pay more for evaluation and treatment.
Both stated that insurance played a role in their decision as well, based on what
coverage was available to pay for visits to the E.D. or physician office.
These preliminary interviews mirrored some of the available data on the role
of perceived physician competence and convenience. However, they also uncovered
a number of previously unreported factors that may play a role in decision-making,
including autonomy, perceived scope of expertise" for each facility, and a complex
schema for evaluation of cost. What emerged was a decidedly multifaceted process
for comparing and prioritizing these factors when making the decision about where to
seek care for non-emergent conditions. Additional qualitative investigation was
necessary to further define these factors, and to learn how they interact with one
another and the individual to result in a decision.

As described above, quantitative studies have identified the profile of
individuals who visit the emergency department. However, no data are available to
answer the foundational research question: what factors do adults who have a choice
consider when selecting where to receive care for non-emergent acute conditions?
While additional quantitative studies could, in fact, shed further light on the research
question, they could not successfully answer the question itself. For instance,
quantitative analysis could be performed on patients who seek care in non-E.D.
settings to compare them with the profile of those who to go to the E.D. However,
while the data might be informative or interesting, they would still fail to describe
how individuals chose where to seek care, and most certainly would be unable to
predict where individuals would seek care. This chapter describes the advantages and
disadvantages of a variety of research designs, and identifies the design utilized in
this study.
Limitations of Quantitative Approaches
Quantitative studies of decision making typically seek to understand the
impact of selected variables (such as cost, facility location, or symptom severity) on
decisions. There are several methodological disadvantages to this approach. First, it
presumes researcher knowledge of the pertinent variables. In his book describing the
limitations of conventional mathematical and economic analysis of the behavior of
farmers behavior, Chambers commented:

In the social sciences and policy, economics dominates, and gives
primacy to mathematical analysis.... Many physical things are
amenable to measurement.... The problem is that the idiosyncratic
attributes of people are, in contrast, difficult to measure: their
individual behavior is unpredictable" (Chambers, 1997).
Quantitative studies typically focus on variables that have been historically
measured such as demographic information, finances, time, diagnoses, and treatments
delivered. There is no assurance that the identified variables are the ONLY that
impact on the decision in question, nor that they are the most influential. In contrast,
qualitative studies focus more on variables as defined by the subjects within their
cultural context. As described by Sobo and DeMunck (comments in parenthesis are
the authors) ... qualitative methods focus on case-oriented (in other words,
individuals) causal processes, whereas quantitative methods focus on variable-
oriented (with the variables identified by the researcher) causality (Sobo, 1998).
Further, in order to reduce the influence of "extraneous" variables, quantitative
studies are performed in controlled settings; elimination of these variables removes
the context from the decision and reduces the applicability of the results in the real
world that is dominated by context and culture (Breslin, 2000; Fjellman, 1976).
Elimination of cultural context also emphasizes the individual as a unitary, rational
decision maker, uninfluenced by external social, economic, or personal constraints; a
variety of studies in the sociology and anthropology fields support the assertion that

real decisions cannot be separated from the broader political, social, and cultural
contexts in which they are made (L.C. Garro, 1998) (Good, 1994).
An additional limitation of quantitative decision research is that it typically
presumes that the individual makes the decision using a rational consideration of the
selected variables in a search for maximum utility. Health care decision research that
studies whether individuals select the ''correct facility based on their symptoms, out-
of-pocket costs, or optimal treatment and follow-up assume that study subjects are
"rational decision makers. Rationality, as discussed by Fjellman in his critique of
decision theory, requires that the decision maker be well informed, able to discern
and discriminate between various options, and able to make decisions based solely on
the analysis of data without influence from emotion, prior experience, other people,
or cultural norms. Further, the rational decision maker must be capable of choosing
the facility that will result in the best possible outcome today and into the future.
After a review of available research, Fjellman concludes:
It appears from a brief look at the experimental literature that none of
the properties assumed in the model (of rational decision making) are
even reasonable approximations of human competence and behavior.
...If we are interested in how people actually act in daily life...we
need to ask descriptive questions about the decision-making process.
(Fjellman, 1976).

This is position that is held by a number of other researchers (C. H. Gladwin,
1976, 1989; Kahneman, 1982a; Tversky, 2002). Finally, the quantitative approach
mathematically describes the probability of behavior across multiple encounters but
does not predict individual behavior; the ability to predict individual choices is
essential for research to have any meaning in the public policy arena (H. Gladwin,
Murtaugh, M 1984a).
Qualitative Methodology
To overcome these limitations this study utilizes a qualitative approach.
Qualitative methods provide descriptions by subjects of their variables and the
processes they use to consider those variables within their own culture and context.
The strength of qualitative research is in evaluating words and behavior to better
describe the variables themselves. As stated by Miles and Huberman in Qualitative
Data Analysis, (qualitative data)... are a source of well-grounded, rich descriptions
and explanations of processes in identifiable local contexts.... Through analysis of
the subjects own words and actions, qualitative research provides a method for the
identification of relevant concepts, processes, and cultural contexts (Breslin, 2000),
the precise process required to answer the question of how patients decide where to
seek care. In addition, qualitative methods facilitate creation of frameworks that
clarify the sequence and effects of events. Again quoting Miles and Huberman,
"With qualitative data one can... see precisely which events led to which
consequences, and derive fruitful explanations.... They help researchers to get

beyond initial conceptions and to generate or revise conceptual frameworks" (Miles,
1994). These advantages are achieved through five specific characteristics of
qualitative research as described by Bogdan and Bilken (Bogdan, 1992): Study
occurs in natural settings (context intact); descriptive method; emphasizes processes
rather than outcome; inductively enables theory to be grounded in data; accurately
describes meanings experienced by subjects themselves. When applied to the study
of health-related behaviors, the purpose of qualitative study is to "seek to understand
what people do when faced with illness and typically attempt to account for actions
taken to deal with illness (L.C. Garro, 1998).
There are many different qualitative research methods. Biography,
phenomenology, grounded theory, ethnography, and case study are all accepted data
collection methods within the qualitative tradition. Each of these methods could
contribute further to the understanding of how individuals decide whether and where
to seek care. For the purposes of this study, specific qualitative decision Modeling
strategies have been developed to better understand the criteria and processes
individuals use to make decisions. These techniques are used in conjunction with a
variety of data collection methods to create models that describe and predict behavior
(Hill, 1998). A model, defined as "a formal representation of the considerations
(individuals involve in their medical decisions) and represents an attempt to describe
the standards of choice in an explicit and testable manner" (L. C. Garro, 1982). From
the perspective of this studys research question, a model offers two distinct

advantages: predictive power and the ability to be validity tested. One of the most
widely utilized qualitative decision Modeling techniques is the Ethnographic
Decision Tree Model. First described by Christina Gladwin in a study of farmers use
of fertilizer in Puebla, Mexico (C. H. Gladwin, 1976). it was later codified into a
monograph describing the method in detail (C. H. Gladwin, 1989).
In the past 30 years the method has been utilized in studies of agricultural
decisions (Darnhofer, Schneeberger, & Freyer, 2005: Fairweather, 1999; C. H.
Gladwin, Peterson, J.S., Mwale, A.C., 2002; McGregor. 2001 ; Murray-Prior, 2001;
Orr, Mwale, & Saiti, 2002), rural economic and market decision making
(Heemskerk, 2002), resource management and consumer recycling behavior
(Calheiros, 2000; G. W. Ryan, Bernard, H.R., 2006), consumer health decisions
(Weller, Ruebush Ii, & Klein, 1997; J. C. Young, 1980); (Oh & Park, 2004, 2006;
Wu, Chao Yu, Yang, & Che, 2005), injection drug users needle sharing (Johnson &
Williams, 1993), and medical decisions by nurses (Foster, Regueira, & Heath, 2006),
traditional birth attendants (Anderson, 2004) and physicians (Breslin, 2000;
Wackerbarth, Tarasenko, Curtis, Joyce, & Haist, 2007).
The name of the method is very descriptive of its attributes. It is ethnographic
because it utilizes ethnographic interviews to learn directly from the decision-makers
the attributes, factors, and constraints used in making their decision. This
ethnographic information is then combined into a decision tree or flowchart that
graphically identifies a series of if-then rules that accurately describe the actions

that derive from the identified factors and constraints. Some researchers suggest that
decision trees created through "manual" qualitative methods are more accurate and
better reflect human decision processes than those created by classification and
regression software (CART) (Darnhofer, et al., 2005).
Ethnographic Decision Tree Modeling
The Ethnographic Decision Tree methodology has a number of distinct
strengths that make it useful for the current study. The process combines in-depth
ethnographic research, using traditional qualitative methods such as interviews and
participant observation with hypothesis testing (L.C. Garro, 1992). Advantages of
the method include an ability to accurately describe the constraints or factors involved
in a decision, and to offer insight about the relationships between the various criteria
included within a decision framework. It may also identify specific decision criteria
that might not be recognized using other qualitative strategies (Breslin, 2000). The
decision trees may be either descriptive or predictive, depending upon the validation
process utilized (Darnhofer, et al., 2005); the method has predicted greater than 80%
of decisions in a broad variety of study environments (Fairweather, 1999; C. H.
Gladwin, 1976; McGregor, 2001 ). When compared with other methods, the
Ethnographic Decision Tree Model has been felt to be more productive than other
methods in understanding complex problems and recommending appropriate
solutions (Calheiros, 2000). A unique strength of the method is the ability to test its

accuracy on other populations using conventional quantitative methods (Bharwani,
2006; Calheiros, 2000; C. H. Gladwin, 1976, 1989)
Decision models, and the Ethnographic Decision Tree Model method itself, do
have a number of weaknesses. Subjects describing a specific decision after it was
made are susceptible to post-hoc rationalizations (L.C. Garro, 1998). Because the
model is based, at least initially, on a typically small, single population surrounding a
specific decision, it may have limitations in applicability to other locations,
populations, or time frames that may have different environmental and cultural
contexts. In addition, since the model ultimately simplifies complex, culturally
enmeshed decisions into a series of'yes/no decision points, it may fail to account
for individual personality factors or unrecognized environmental cues (Darnhofer, et
al 2005).
A number of studies have utilized the Ethnographic Decision Tree Model
method to describe health-related decisions by health practitioners and consumers.
Anderson et al. utilized ethnographic interviews to learn how traditional birth
attendants (Parteras) manage complicated childbirths. Through in-depth interviews of
six Parteras practicing in rural Mexico, the descriptive study identified specific
decision points that led to a variety of interventions including transport to the local
hospital. Although the authors described using Gladwins decision tree method, they
did not create a decision tree, nor validate their findings internally or externally
(Anderson, 2004). In a study also related to rural management of childbirth

complications, Foster et al. interviewed eight auxiliary nurses in a Dominican
Republic maternity ward to learn what criteria they use to assess post-partum
bleeding. Following creation of a decision tree based on the interview findings, the
authors validated the model through a survey administered to 16 nurses who did not
participate in the initial interviews; 94% of the survey responses were consistent with
the original decision model. The authors also validated the model through
presentation to hospital personnel for additional input, and through participant
observation within the hospital (Foster, et al., 2006).
Breslin and a team that included Christina Gladwin created an Ethnographic
Decision Tree Model to describe how clinicians in Toronto make patient referrals into
two distinct outpatient substance abuse treatment programs. The model described 16
separate decision points, and was created following interviews of eleven clinicians.
The subjects were asked to broadly describe their referral guidelines, and to describe
one recent case that they had referred to one of the programs, and then a different
case that they had referred to the other program; the initial model accurately
described the referral of 90% of cases. During a second study phase model validation
was performed by a second interview of the same clinicians in which they were asked
to describe two additional referral cases, one to each of the programs. In the
validation sample the model accurately predicted the referral of 85% of cases.
A number of studies have utilized the Ethnographic Decision Tree Modeling
strategy to predict health consumer decisions. Wu and her colleagues created a

decision model to predict how women in Taiwan make the decision whether or not to
have a hysterectomy. Ethnographic interviews were performed on 14 women whose
physicians had recommended they undergo a hysterectomy. None of the subjects had
undergone the procedure at the time of the interview; 12 stated that they would go
through with the hysterectomy, while two stated they would decline. The model
identified 13 factors that influenced the decision. Validation was performed through
the interview of 18 additional women whose physicians had recommended
hysterectomy; the model accurately predicted 90% of the choices in the second group
(Wu, et al 2005).
A similar strategy was employed by Oh and colleagues in a study of treatment
seeking behaviors of Korean cancer patients. Twenty-nine cancer patients were
interviewed to identify the five factors that constituted the decision model. Decision
tree testing was performed using a combination of surveys and interviews involving
165 cancer patients. Predictability was calculated separately between two different
groups of patients: those who visited their doctor within one month of detecting
symptoms, and those who visited their doctor more than one month of detecting
symptoms. Overall predictive accuracy was 90.3%. In addition to the validation
processes utilized in the other studies described, the authors also used regression
analysis to identify factors related to seeking treatment immediately; serious
symptoms, previous experience visiting a doctor, and suggestions from others were
all statistically significant at the 5% level (Oh & Park, 2004).

Young and Garro utilized the Ethnographic Decision Tree Modeling strategy
to describe medical decision making in a rural Mexican village (C. Young, Garro, L,
1981). The study included ethnographic data collection of 323 illness episodes
(involving 489 individual treatment choices) experienced among 62 households in
Picataro, Mexico. The researchers identified four primary criteria for making
treatment decisions: seriousness of symptoms, knowledge and experience of a
treatment for the illness being experienced, faith in the efficacy of available folk"
treatment in comparison to more formal medical treatment, and the expense of
treatment alternatives (and the availability of resources to pay). As described in a
paper following the initial publication (J. C. Young, 1980), creation of the decision
model evolved during the analysis of the initial 300 treatment decisions, and was
94.7% accurate in accounting for the individual choices. Validation of the model was
performed through application to the remaining 144 treatment decisions; accuracy of
prediction during validation was 84%, for an overall accuracy of 91.2% for the entire
Weller et al. compared the ability of two different qualitative methods to
predict the treatment seeking behavior of individuals in rural Guatemala (Weller, et
al., 1997). The study compared the effectiveness of Ethnographic Decision Tree
Modeling with the Sociobehavioral model. While both methods utilize traditional
ethnographic data collection methods, the Sociobehavioral model differs from the
Ethnographic Decision Tree method in the structure of interviews and surveys. In

contrast to the open-ended, subject-focused questioning described by Gladwin,
questions used in the Sociobehavioral model are based on theoretically pre-
determined factors: predisposing (age, gender, pre-existing health), enabling (access
to care, financial resources), need (severity of symptoms), and treatment (self-care,
pharmacy, health post, hospital, and folk curer). Two interview surveys (25/20
families) were performed as part of the Sociobehavioral study. Interviews were
conducted with 15 families to construct the Ethnographic Decision Model, which was
validated using vignettes (based on the decision factors) that were read to an
additional 25 families. The researchers found that the Decision Model independently
identified decision factors that were concordant with the variables identified through
the Sociobehavioral study. Both studies identified perceived severity as the most
important single factor guiding health seeking behavior. The researchers concluded
that the Decision Tree model was useful in describing how an individual might
behave, in contrast to the population-based prediction of the Sociobehavioral model.
The authors drew the following conclusion concerning the Sociobehavioral model:
An advantage of the SBM is that it offers a theoretical framework for choosing
variables that affect choice...(and)...would have provided important information
regarding the selection of variables for the (Ethnographic) decision model..This
suggested predetermination of variables is in direct conflict with one of the
advantages of the Ethnographic Decision Tree Model method as stated by Gladwin:

Ethnographic Decision Tree Modeling starts from the assumption that
the decision makers themselves are the experts on how they make the
decisions they make....Discovering the insider's world from the
insider's point of view is a far different goal from that of collecting
data about people and testing a model based on the outsider's view.
(C. H. Gladwin, 1989).
Garro contrasted the use of Ethnographic Decision Tree Modeling in studies
of decision making strategies of subjects in two distinctly different locations,
Tarascan natives in Pichataro, Mexico (previously described) and members of the
Anishinaabe community in Manitoba, Canada (L.C. Garro, 1998; Linda C. Garro,
1998). She made several important distinctions about the environment and culture of
the two populations. While the Pichatarenos experience significant limitations of
both access to care and availability of resources to pay, the Canadian sample has easy
access to care andbecause of the socialized nature of the Canadian health care
system virtually no cost constraints. An additional variation was found in the
"gatekeeping function of primary caregivers in Canada, a barrier to care that is not
present in Pichataro. There was also an important cultural difference that impacted
the decision Modeling study. Within the Anishinaabe community there is a strong
belief that illnesses and injuries can be "linked and may reflect the presence of "bad
medicine. This belief changes the perception of some acute symptoms from being
isolated physical events to chronic spiritual ones and, as a result, alters the decision

process for seeking care. While the decision tree Modeling methodology was useful
in identifying what Garro referred to as the "decision making perspective" of both
populations, she found that it was limited in its "decision Modeling" capacity in the
Canadian study because of this cultural difference in the meaning of symptoms. Her
conclusion was an important one in relationship to the applicability of Ethnographic
Decision Tree Modeling:
While research in both sites affirmed the broad applicability of a
decision-making perspective to the understanding of actions taken in
response to illness, I remain open to the possibility that this approach
may not be as productive in other cultural settings, or that it may not
even be applicable (Linda C. Garro, 1998).
The studies described demonstrate a broad applicability of the Ethnographic
Decision Tree Modeling method in a wide variety of health settings. A summary of
studies that have utilized this method can be found in Appendix B. However, in the
current study and elsewhere it remains essential to evaluate both the validity of the
results and the appropriateness of the method itself in light of the cultural
environment and health perceptions that are present in the community under
Ethnographic Decision Tree Modeling Process
Based on the strengths described above, the current study utilizes the
Ethnographic Decision Tree Modeling method to identify the factors individuals

consider when choosing where to seek care. The process for creating and validating
the decision model can be broken into five steps (C. H. Gladwin, 1976):
Gather data and identify decision constraints
Create a composite decision flow chart (tree)
Test the internal validity of the decision tree
Identify ordering aspects and relationships of the identified decision
Test the external validity of the decision tree
Gather Data
Using conventional ethnographic data collection tools such as participant
observation and in-depth interviews, data are gathered from 20-30 decision makers
about the decision. Interviews are performed using open-ended approaches to
discover what choices they see as available to them, what decision criteria they use to
choose among them, and what "rules of thumb guide their decision process. Data
must be carefully recorded to assure that the decision criteria identified are in the
subject's words from their perspective and NOT structured in a manner consistent
with the researchers beliefs and biases. The questioning process becomes iterative
and cyclical; hypotheses regarding decision criteria that are generated in early
interviews lead to evolution of interview questions in later interviews to test the

Create the Decision Tree
Using the decision points identified in the first step, and using the language
and categories described by the informants, a decision flow chart is created. Flow
charts may be created using either direct or indirect methods. In the "direct method"
a flow chart is created for each informant, and then a composite tree is creating using
the attributes identified in the individual trees. In the "indirect method" the
composite flow chart is created as the interviews progress (C. H. Gladwin, 1989; Oh
& Park, 2004). Creation of the composite model using either strategy requires
patience and a compulsion to return to the data frequently to identify patterns and to
assure that patterns that have been created have their foundation in the informants'
words. Strategies that may be employed in creating the model(s) include the creation
of vignettes from subject data that incorporate the various alternatives that could
become decision criteria. Another straightforward strategy is to examine any cases
that dont fit the proposed decision tree, and then modify it until they fit. This
process is continued, in combination with the removal (pruning) of branches that are
only utilized by one to two individuals, until a tree is created that fits the majority of
subjects. More sophisticated approaches involve the use of artificial intelligence,
classification and regression tree analysis (CART) (Oh & Park, 2004; G. W. Ryan,
Bernard, H.R., 2006).

Test Internal Validity
The internal validity of the composite decision tree is tested by comparing it
to the decisions made by the initial 20-30 decision makers. Internal validity is
reported as the percentage of actual decisions that are consistent with the model.
Because the tree itself was constructed based on the decision criteria defined by these
individuals, the internal validity should be fairly highusually over 90%-95%. It is
useful to identify cases/decisions that are not in agreement with the model to
determine if they represent is single decision factor that may not be adequately
described by the model. If this is the case, or if the internal validity is lower than
90%, it is probably necessary to continue revising the composite tree (C. H. Gladwin,
Identify Aspects and Relationships Among the Decision Criteria
The composite decision tree provides useful information concerning the
decision criteria themselves, but offers no insight regarding the processes and
rationale behind the criteriaone of the advantages of the Ethnographic Decision
Tree Modeling method. Using the themes and words used in the participant
interviews, the relationships between the various decision points are examined. Are
they sequential? Do some criteria depend upon others? Are the relationships
between certain criteria? Can the criteria be ordered based on cost, risk, or impact on
the subject? The goal is to identify the themes that underlie the decision criteria, and
to understand how the themes interrelate.

Test the External Validity of the Decision Tree
The process of testing internal validity demonstrates the applicability of the
model to the initial group of subjects. However, it offers no insight regarding the
applicability of the described decision criteria to other individuals within the same
population, or to other populations. External testing can be done in a variety of ways.
If the initial subjects are still available, they can be interviewed again and asked more
closed-ended questions that are focused on validating the decision model. A similar
strategy could be exercised by creating a survey that asks "yes/no question(s) for
each decision criteria, and administering the survey to the original subjects. When
possible it is preferable to perform external validity on a new group of subjects from a
similar population using either the interview or survey format. As with internal
validity, cases or decisions that are not consistent with the model should be analyzed
to ascertain whether there is a single subject group or decision criteria that seem to be
poorly described by the model. This may become a limitation of the model, or lead to
further revisions to better describe the behavior of the informants. External validity is
reported as a percent of decisions that are consistent with the model. Although
Gladwin initially recommended that 90% be the minimum acceptable external
validity (C. H. Gladwin, 1976, 1989), a number of respected studies have reported
validity of less than 90% (H. Gladwin, Murtaugh, M., 1984b; G. W. Ryan, Bernard,
H.R., 2006), and at least one author has suggested that the limited number of studies

utilizing the methodology may make establishment of a single reference for validity
unrealistic (L.C. Garro, 1992).
Ethnographic fieldwork assumes that decision-makers are the experts in the
process being studied. In this study, that assumption is manifested by the use of
qualitative ethnographic interviews to determine the factors used by individuals in
deciding whether and where to seek care. While quantitative research has proposed a
number of factors underlying this decision (these were presented above), it would be
an ethnocentric mistake to assume that they are the same factors used by the subjects
within this study. Building a decision tree model based on faulty assumptions of key
factors could perpetuate specious facts and create a final product that cannot
accurately predict behavior. An important attribute of the Ethnographic Decision
Tree Modeling process is its insistence on formal, testable decision process
descriptions. The hierarchical decision tree that is created is testable, and its ability to
accurately predict behavior quantified. This feature not only improves the quality of
the final product, but also takes the study results beyond mere conjecture to
something that can be replicated, tested, and applied to other populations. In short,
the process has inherent rigor that improves both reliability and its acceptability to
other sciences and disciplines.

The study utilized a qualitative interview process to determine the factors
consumers utilize to decide where to seek care. The study itself was broken into two
distinct phases. The first was the initial data gathering phase that included interviews
of study participants. The second was a validation phase that included creation and
distribution of a survey that was constructed based on the findings of the initial phase.
Settings and Sample
Study Setting
The study sample included insured individuals who received care for a non-
emergent acute medical problem at a healthcare facility. To reduce variability that
might be caused by different health problems the decision was made to only consider
non-traumatic complaints, and to focus on a single, common constellation of
complaints related to upper respiratory symptoms. To assure a diverse sample
participants were recruited from three sites selected for the study in the center of a
large city in a rocky mountain state; all three sites were within 6 miles of the others.
Participating sites included an emergency department, urgent care clinic, and private
physician practice. For the purposes of this study, here are the definitions of these
Emergency department: a hospital-based, accredited, 24/7 facility staffed by board
certified or board eligible emergency physicians.

Urgent care clinic: a freestanding facility (not attached to a hospital) that advertised
itself as an urgent care facility," staffed by one to two physicians at a time. The
clinic does not have multi-specialty physicians available on-site, and does not provide
inpatient capability.
Private Physician Practice: a freestanding practice (not attached to a hospital) staffed
by several physicians who represent a single specialty (such as family practice or
internal medicine). The practice does not have multi-specialty physicians available
within the practice, and does not provide inpatient capability. It also does not have its
own laboratory or radiology capability.
Participating Facilities
Three facilities agreed to participate in the study. To assure confidentiality
the facility names have been changed:
Emergency Department: Good Samaritan Hospital (GSH) is a private, non-
profit, JCAHO accredited 593 bed acute care hospital located near the geographic
center of a large urban center in a Rocky Mountain state. The emergency department
is open 24 hours/day, 365 days/year and sees approximately 43,000 patients per year.
Urgent Care Center: PromptCare Clinic (PC) is an independent urgent care
center located in a suburb 5.3 miles due west of GSH. The facility is owned by a
group of physicians, and experienced approximately 20,000 visits in 2004. The
facility is open from 8 am to 8 pm, 365 days/year, and has included a laboratory & x-
ray facility on-site.

Physician Office: Waterbridge Family Medicine (WFM) is a small two-
physician group practice located in a suburb 2.7 miles northwest of GSH and 3.6
miles northeast of PC. The office is open Monday through Friday from 8 am to 4 pm
and provides family practice and internal medicine. The practice sees approximately
40 patients each day and does not have laboratory or x-ray capabilities on-site.
The study research design and protocol was approved by the Institutional Review
Board of Good Samaritan Hospital. More information about the institutional review
can be found in the IRB application in Appendix C.
Sampling Frame and Participant Recruitment
Interview Phase
Participating facilities agreed to have the principal investigator present in their
facility to speak with potential study participants during their visit to the facility. No
medical records or patient identifying information (other than the patient's name at
introduction) were shared with the principal investigator. The principal investigator
was present at GSH on ten dates and PC on nine dates, throughout their hours of
operation to identify and meet with potential subjects. Because of their smaller
facility and lower patient volume, staff from WFM identified subjects that appeared
to meet the inclusion criteria from their daily appointments and notified the principal
investigator who went to the facility during the appointment time.
Inclusion/exclusion criteria were established to support the study aims: to
identify the factors that adults who have a choice consider when seeking care for non-

emergent health problems. Only adults were included, and choice" was determined
by assuring that participants had some form of health insurance. Individuals who
were brought to the facility by ambulance were excluded because of the influence
paramedics may have had on their chosen destination, and because ambulance
protocols typically prohibit transport to urgent care facilities or physician offices. To
stay consistent with the study's aims and assure patient safety, individuals who were
perceived to have a health problem that required immediate intervention were
excluded from the study. Based on these principles the inclusion criteria for study
participants included:
21 years of age or older
Coverage by medical health insurance (confirmed by valid insurance
identification card)
Seen in the facility for an acute, non- routine appointment (or without an
Sought care for a non-traumatic acute medical complaint associated with
cough or upper respiratory symptoms (such as runny nose).
Participants were excluded from the study if they were admitted to the facility
by ambulance.

Participants were excluded from the study if their diagnosis was felt to be
emergent", as defined as requiring treatment in < 15 minutes. This
determination was made by the treating physician and confirmed by the
principal investigator.
Eight participants who met inclusion criteria were recruited from each facility.
All study participants received written and verbal information about the study and
signed the informed consent prior to enrollment. More information about participant
safety and institutional review can be found in the IRB application in Appendix C.
Following completion of informed consent, each participant completed a Subject
Information Form that included the following information (a copy of the Subject
Information Form can be found in Appendix D.):
Demographic information including age, race, gender, level of education,
income range.
Insurance information, including name of insurer, amount of copay (or
awareness of copay), who pays for insurance (i.e. employer, self)
Past medical information, including brief past medical history, nature of
relationship (if any) with primary care provider (PCP), previous experience
with non-emergent acute medical conditions.
Validation Phase
Validation of the Ethnographic Decision Tree was performed through
distribution of a survey that was developed based on the decision criteria identified in

the composite ethnographic decision tree. For consistency with the initial study and
convenience, subjects were recruited from the same three sites (Good Samaritan's
ED, PromptCare, and Waterbridge Family Medicine). To test the model on a broader
group of patients, subjects were eligible for participation with any acute complaint
(not just upper respiratory symptoms). Surveys were distributed to a convenience
sample of individuals who met the following inclusion criteria:
Age 2 lor older
Presented to the facility with any acute complaint (medical or trauma), and not
for a routine checkup
Do not enter the facility via ambulance
Medically stable enough to take the survey (in the opinion of the caregiver
administering the survey)
Able (self or family member) to read and understand English
Unexpectedly, the ownership of PromptCare changed between the interview
and survey stages and the new owners were unwilling to have surveys distributed to
their patients. Further information about the development of the validation survey
can be found in the Data Analysis section below. The validation process and survey
were approved by the GSF1 Institutional Review Board on September 11, 2008, and
can be found in Appendix F.

Data Collection and Management
Interview Phase
Data sheets from each participant were coded by the researcher to identify the
specific facility where the patient was recruited and the patient's chief complaint.
A qualitative interview (average length 35-45 minutes) was performed with each
participant to explore the factors behind the decision to seek care at the participating
facility. Participants enrolled at GSH and WFM were interviewed at the time of the
healthcare encounter within the facility; interviews were completed for all individuals
enrolled in these settings. Because of a lack of appropriate interview space, subjects
enrolled at PC were scheduled at the time of the encounter for a phone interview to
take place within the subsequent 1-2 days. Participants were paid $25 following
completion of the interview. A question guide was used as the framework of the
interviews; issues were explored beyond the question guide as necessary to assure
that the individual's decision factors were fully understood. The question guide can
be found in Appendix E.
Interviews were taped to assure accuracy, and then transcribed by the
principal investigator using HyperTranscribe ("HyperTranscribe," 1997), a computer-
based transcription program that operates directly from electronic audio files. Audio
files were maintained for archival purposes. To assure the accuracy and cultural
consistency of the data, content transcription was performed according to the

Qualitative Data Preparation and Transcription Protocol recommended by McLellan
et al. (McLellan, MacQueen, & Neidig, 2003). Key features of the protocol include:
Audiotapes were transcribed verbatim
Nonverbal sounds (such as laughter) were included and typed in parenthesis
Transcription included the grammar, sentence structure and word use chosen
by the subject with no clean-up"
Transcription faithfully recorded blended words (such as kinda") as used by
the participant
Filler words such as urn" or yeah were transcribed as used by the
All transcripts were reviewed by the primary investigator for accuracy, spell-
checked (with the exception of the protocol described above), and saved as rich text
files for upload into Atlas ti (Muhr, 2004) for qualitative analysis. Qualitative
analysis was performed on all interviews; further description of the process and
findings can be found below.
Validation Phase
The survey itself had no indication of where the individual received care;
rather, all surveys were numbered and a key was developed that identified the care
site for each numbered survey. Surveys were distributed to subjects meeting the
inclusion criteria at GSH and WFM between October 20 and November 24, 2008.

Data Analysis
To facilitate achievement of the study aims raw interview data were analyzed
following the tenets of grounded theory. Grounded theory techniques enable the
identification of themes from spoken and written words, and the linking of those
themes through the creation of models and theory (Bernard, 2002). Analysis was
performed in two stages. The first stage, thematic analysis, was performed to identify
the factors and constraints that adults consider when choosing where to receive care.
In addition to addressing the first study aim, thematic analysis also provided the
foundation for the second stage, creation and validation of the Ethnographic Decision
Tree Model.
Thematic Analysis
Thematic analysis, also known as domain analysis, was performed using an
inductive approach because it enables the discovery of the themes articulated by the
subjects themselves (Bernard, 2002; Fereday, 2006; Lewins, 2007; Pope, Ziebland, &
Mays, 2000; G. W. Ryan & Bernard, 2003), takes advantage of the rich information
gained through participant interview, and offers the greatest opportunity to hear the
participants voices. Analysis was performed as described by Ryan and Bernard (G.
W. Ryan & Bernard, 2003) using four main strategies: word analysis, careful reading
of text, and pawing, cutting, and sorting. The use of these strategies in the study is
briefly discussed below.

Word analysis was performed by mechanical counting of word usage within
subject interviews to identify frequently used words; this process offered one source
of information about topics of importance to participants. As Ryan and Bernard
noted, words that occur a lot are often seen as being salient in the minds of
respondents" (G. W. Ryan & Bernard, 2003). Once the frequently-used words were
identified, their contexts within the interview transcripts were categorized to better
understand the meaning of the word uses within the participant's experience. This
process is described as key words in context" or KWIC" by Miles and Huberman
(Miles, 1994). Frequently used words and their contextual content were used to
inform subsequent steps in the thematic analysis.
Following this abstract review of word usage and context, the transcripts were
carefully reviewed again to identify specific passages that vividly described the
participants feelings and experience related to seeking care. Initially these quotes
were simply identified with no attempt to identify themes or codes. Comparisons
were made of statements made by subjects, and shared and contrasting perspectives
were labeled using memos as described by Pope, Ryan and Bernard, Spradley, and
Lewins and Silver (Lewins, 2007; Pope, et al., 2000; G. W. Ryan & Bernard, 2003;
Spradley, 1979). The goal was to simply capture their important statements. The
next step involved returning to the quotes with a goal of identifying themes that were
common among the participants; in many cases these themes were consistent with
those identified during the word analysis and became the initial codes. This process

of repeatedly reviewing transcripts, marking statements of interest, comparing and
contrasting quotes among participants, and sorting similar statements into thematic
groups has been described as pawing (highlighting and marking texts), cutting
(pulling important statements from the texts) and sorting (placing similar statements
from different subjects into groups that become themes) (G. W. Ryan & Bernard,
2003). These processes resulted in the creation of the initial inductive themes.
Following identification of tentative codes the transcripts were reviewed again in an
almost deductive fashion to identify any additional participant statements that
pertained to the already-identified themes. When found, these quotations were saved
and labeled with the applicable code. In addition to identifying the utility and
occurrence of the emerging themes, this process also assured that the themes
originated with the subjects themselves rather than the researcher's expectations. The
quotations identified during this process provided an additional context of reality
within the subjects' culture for the themes that further explicated their meaning and
The final step in the thematic analysis involved the identification of
relationships between the themes, unique subject characteristics, and/or other themes.
Differences in the frequency of word use or specific codes among subjects with
differing age groups, gender, educational backgrounds, or socioeconomic groups
were explored. While not statistically significant, these differences provided further
context for thematic understanding, and offered suggestions for future research.

During this stage relationships between the themes themselves were also identified,
and provided useful information about potential relationships such as cause and
effect, or specific steps in the decision process.
The product of this inductive thematic evaluation was a robust list of
interrelated themes that helped to describe the factors that participants considered
when deciding where to seek care. Good descriptions of this process can be found in
The Ethnographic Interview by Spradley (Spradley, 1979) and an excellent article by
Atkinson and Hap (Atkinson & Hap, 1996).
Ethnographic Decision Tree Modeling
The creation of a list of interrelated themes through thematic analysis was a
valuable process that identified the factors and constraints that individuals consider
when deciding where to seek care. However, it was primarily descriptive in nature
and offered little explanatory power and no predictive potential. Creation of
Ethnographic Decision Tree Models facilitated a better understanding of how and
why participants make actual decisions, information that has the potential to predict
actions and provide an objective tool to evaluate the impact of an intervention (such
as changing a co-pay to encourage patients to seek care in an alternative location).
An additional advantage is that an Ethnographic Decision Tree can be concretely
tested in subsequent investigations (C. H. Gladwin, 1989). Based on the data
gathered during the thematic analysis, Ethnographic Decision Trees were developed
and validated through the steps described below.

Creation of Individual and Composite Decision Trees
Because of the iterative nature of both the interviews themselves and the
creation of the decision tree, the "direct" method was chosen. A decision tree was
created following each interview; these individual decision trees identified potential
decision points for investigation and testing in subsequent interviews as described in
the literature (C. H. Gladwin, 1989; Oh & Park, 2004). With each interview
additional decision points were identified and repeated processes were clarified.
After approximately 18 interviews no additional decision points were identified
(saturation), although the frequency and patterns of decision processes continued to
be placed in sharper focus.
Once individual decision trees were completed on a significant number of
participants they were compared and contrasted to the themes and relationships
identified during the thematic analysis. This process provided necessary context to
the decision models, and identified relationships between the themes and decision
steps. At this time efforts began to combine features of the individual trees into a
composite model. The process was somewhat of a hybrid between the two basic
methods described by Gladwin, ad hoc and rationale (C. H. Gladwin, 1989). When
they appeared obvious and consistent, similar "branches from individual trees were
combined according to the ad hoc method. At the same time, the interview text was
reviewed to better understand the reasons behind the various decision criteria. These
reasons were sometimes consistent with already identified themes; in other cases they

offered new insights into the rationale behind subject decisions. These reasons were
compared among study subjects to identify common rationale. Repeated
combinations of ad hoc and underlying rationale strategies resulted in the iterative
construction of a series of draft composite decision models, and ultimately led to a
single model that appeared to predict the decision behavior of the study subjects. The
composite model honored the themes identified during thematic analysis and was
compatible with a majority of the decision models created for each subject.
Internal Validity Testing
Internal validity testing describes the accuracy of the composite model in
predicting the decisions of the individuals who were interviewed. To test for internal
validity each subject was run through the composite model to determine whether it
accurately predicted their chosen treatment site. During this process the relationships
between various decision constraints became obvious, and several final modifications
and simplifications were made to the composite model to better reflect these
relationships and sequences. The resulting Ethnographic Decision Tree Model
represented the decision making strategies for the limited number of individuals who
were interviewed, and its internal validity was calculated as the percentage of
decisions that were accurately predicted by the model.
External Validity Testing
The ability of the model to predict behavior of additional individuals can only
be posited following external validation testing. The external validation process was