Sociodemographic characteristics and survey response bias, Brinkman et al.
19 Patient Experience Journal, Volume 8, Issue 1 – 2021
within pediatrics.
9
Since survey participation differences
can affect the generalizability and usefulness of PFE
surveys results,
1
such differences are critical to understand
participation biases. Such data is lacking especially in the
context of pediatric healthcare, leaving a gap in the
understanding of potential participation biases in PFE data
sets. Thus, we sought to enumerate response, completion,
and comment rates to our PFE survey across key
sociodemographic characteristics.
Patients and Methods
PFE Survey Methodology
This retrospective study took place at Cincinnati
Children’s Hospital Medical Center (CCHMC) using data
collected between January 1, 2019 and July 31, 2019.
CCHMC is a large, urban, free-standing, academic, acute
care, children’s hospital. For patients contributing more
than one visit during the study period, one visit was
selected at random. This study analyzed 77,627 encounters
from ambulatory medical and surgical specialty divisions
where surveys where administered. The study was
reviewed and deemed exempt by the CCHMC Institutional
Review Board.
PFE surveys are administered following clinical encounters
using the NRC Health Real-Time survey platform.
10
Families are automatically contacted within 24-48 hours of
an ambulatory visit requesting completion of the PFE
survey. This outreach is initially via email, using an email
address provided at the time of clinical encounter. If no
response is received within 24 hours of the email, a second
outreach via an interactive voice response (IVR) phone call
is attempted. If there is still no response following an
additional 24 hours, a second email is sent. If there is no
email on file, or the email is a CCHMC institutional email,
all three outreach attempts are made via an IVR phone
call. Families have up to 14 days to respond and complete
the survey before being identified as a survey non-
responder.
Survey Exclusion/Inclusion Methodology
Several exclusion criteria exist as part of the NRC Real-
Time platform based on demographic, diagnostic, and
prior utilization factors. To minimize survey fatigue, no
survey is attempted if a patient received a survey from any
other clinical encounter within three days of the
ambulatory encounter or from the same ambulatory
division within the previous 30 days. The survey is only
administered in English or Spanish based on a patient’s
self-reported primary language. Additional exclusions exist
for encounters connected to an international address,
those with incomplete or erroneous contact information,
and patients in custody of the state or county (<1% of
encounters). Finally, to prevent accidental disclosure and
to protect patient privacy, patients evaluated for a set of
“sensitive diagnoses” chosen a priori (e.g., pregnancy,
suspected or verified abuse, and visits related to sexually
transmitted infections) were excluded.
PFE Survey Elements
Each survey consisted of 14 questions that assess common
PFE priorities, including access to care, communication
with the care team, continuity and coordination of care,
information and education, respect, environment of care,
and overall rating of experience. Responses to the
questions were anchored to either a three-point or four-
point scale, or a 10-point Likert scale for the overall rating
of the provider question. Questions and response options
are in Figure 1. After these questions, respondents have
the option of leaving a free response comment.
Outcomes and Sociodemographic Variables
The outcomes of interest were PFE survey response,
completion, and comment rates. A survey response was
defined as answering at least one of the Likert-based
questions. Completion was defined as answering all the
Likert-based multiple response questions on the survey.
Commenting was defined as inclusion of any free-text
comment.
The patient level sociodemographic variables included
were race, ethnicity, primary language, and socioeconomic
deprivation. These sociodemographic characteristics were
extracted from the electronic medical record. Race was
categorized as White, Black or African American,
Multiracial, other, or unknown. Other consisted of Asian,
Hispanic, and other. Ethnicity was categorized as Hispanic
or non-Hispanic, while language was segmented into
English, Spanish, or Other/Unknown. Race, ethnicity, and
preferred language are self-reported by patients and
families during the clinical encounter and entered into the
electronic health record. Patients can self-identify as both
Hispanic race and Hispanic ethnicity.
Socioeconomic deprivation was estimated using the
Deprivation Index (DPI) measured at the census tract
level. This measure is used to approximate respondent
socioeconomic status and characterize the context in
which they live. The DPI is calculated from variables
present within the U.S. Census American Community
Survey (ACS) such as poverty, educational attainment, and
access to health insurance.
11
For this analysis, we grouped
respondents into quartiles by their DPI scores with higher
DPI indicating higher deprivation. The DPI was obtained
by extracting the patient street address from the electronic
health record at the time of the encounter which was then
geocoded and connected to the corresponding census tract
geography. Once the census tract was identified, it was
then linked to the DPI value.