Using

Geographic
Information Systems (GIS) to Assess Outcome
Disparities in Patients with Type 2 Diabetes an
!yper"ipiemia
Estella M. Geraghty, MD, MS, MPH/CPH, FACP, Thomas Balsbaugh, MD,
Jim uo!o, MD, a"# Sa"$ee! Ta"#o", MBBS, MD
Objectives: Geographic information systems (GIS) tools can help expand our understanding of dispari-
ties in health outcomes within a community. The purpose of this project was (1) to demonstrate the
methods to lin a disease management registry with a GIS mapping and analysis program! (") to ad-
dress the challenges that occur when performing this lin! and (#) to analy$e the outcome disparities
resulting from this assessment tool in a population of patients with type " dia%etes mellitus.
Methods: &e used registry data deri'ed from the (ni'ersity of )alifornia *a'is +ealth System,s elec-
tronic medical record system to identify patients with dia%etes mellitus from a networ of 1# primary
care clinics in the greater Sacramento area. This information was con'erted to a data%ase file for use in
the GIS software. Geocoding was performed and after excluding those who had unnown home ad-
dresses we matched -."- uni/ue patient records with their respecti'e home addresses.
Socioeconomic and demographic data were o%tained from the Geolytics! Inc. (0ast
1runswic! 23)! a pro'ider of (S )ensus 1ureau data! with "44- estimates and projections.
5atient! socioeconomic! and demographic data were then joined to a single data%ase. &e
conducted regression analysis assessing 61c le'el %ased on each patient,s demographic and
la%oratory characteristics and their neigh%orhood characteristics (socioeconomic status 7S0S8
/uintile). Similar analysis was done for low-density li-poprotein cholesterol.
Results: 6fter excluding ineligi%le patients! the data from 9"-- patients were analy$ed. The
most nota%le findings were as follows: There was! there was found an association %etween
neigh%orhood S0S and 61c. S0S was not associated with low-density lipoprotein control.
Conclusion: GIS methodology can assist primary care physicians and pro'ide guidance for disease
management programs. It can also help health systems in their mission to impro'e the health of a com-
munity. ;ur analysis found that neigh%orhood S0S was a %arrier to optimal glucose control %ut not to
lipid control. This research pro'ides an example of a useful application of GIS analyses applied to large
data sets now a'aila%le in electronic medical records. (3 6m 1oard <am =ed "414>"#:-- ?@A.)
Type 2 diabetes mellitus (DM) is one of the most
common chronic medical conditions in the United
States. The Centers for Disease Control and Pre-
ention estimates that there are appro!imately "#
million people $ith DM.
"
%nd-or&an complica-
This article $as e!ternally peer reie$ed.
Submitted "' (une 2))*+ reised "" ,oember 2))*+
ac-cepted "- ,oember 2))*.
From the Department of .nternal Medicine (%M/)0 the
Department of 1amily and Community Medicine (T20 (,)0
and the 3ealth .nformatics Pro&ram (ST)0 the Uniersity of
California Dais School of Medicine0 Sacramento.
Funding: none.
Conflict of interest: none declared.
Corresponding author:
Thomas 2alsbau&h0 MD0
Depart-ment of 1amily and
Community Medicine0
Uniersity of California
Dais School of Medicine0
45') 6 Street0 Suite 2-))0
Sacramento0 C7 *85"# (%-
mail9 thomas.balsbau&h:
ucdmc.ucdais.edu).
tions from DM are a
substantial source of
morbid-ity and mortality.
Disparities for aderse
outcomes associated
$ith DM are $ell
documented.
2
1actors
that are associated $ith
these disparities include
lan&ua&e barriers0
inade;uate access to
care0 lo$
socioeconomic status0
and suboptimal self-
care be-haiors.
-
Si&nificant disparities e!ist amon& racial< ethnic
minorities in DM health outcomes and ;ual-ity of
care0 includin& poorer measures of &lycemic control
and hi&her rates of end-or&an complica-tions.
4
%fforts to improe outcomes of patients $ith
chronic conditions has included implemention of the
Chronic Care Model
(CCM).
8
The CCM $as
deeloped to identify the
essential elements of a
health care system that
encoura&e hi&h-;uality
chronic disease care.
These elements include
the
55 361<=
(anuary=
1ebruary 2)") >ol. 2- ,o. "
http9<<$$$.?abfm.or&
community0 the health system0 self-mana&ement
support0 deliery system desi&n0 decision support0
and clinical information systems.
8
Clinical infor-
mation systems may sere as reminders to help
teams comply $ith practice &uidelines0 as report
cards to proiders about measures of care0 and as
re&istries for plannin& indiidual patient care and
conductin& population-based care.
'
@hen con-
ductin& population-based care0 &eo&raphic infor-
mation systems (/.S) tools e!pand our
understand-in& of disparities in health outcomes
$ithin a community.
#
The purpose of this pro?ect $as --fold. @e
$anted to demonstrate the methods to linA a dis-
ease mana&ement re&istry $ith a /.S mappin&
and analysis pro&ram+ to address the challen&es
that occur in performin& this linA+ and to describe
the contributin& factors associated $ith differences
in outcomes in a population of patients $ith DM0
specifically the factors associated $ith optimal &lu-
cose control and treatment of hyperlipidemia.
=ethods
The Chronic Disease Mana&ement Pro&ram at the
Uniersity of California Dais 3ealth System
(UCD3S) is a system-$ide effort to improe the
;uality of care for patients $ith chronic conditions.
The CCM sered as the paradi&m by $hich the
pro&ram $as structured.
8
The Chronic Disease
Mana&ement Pro&ram maintains a re&istry of all
patients $ith DM $ho receie care in the UCD3S.
University of California Davis Medical Center
and Primary Care Network
The patient re&istry data from this study included
all patients $ith DM $ho $ere seen $ithin the
clinics at the Sacramento-based medical center
and our re&ional primary care net$orA ("- clinics
rep-resented in this analysis). Clinic addresses
$ere &eocoded $ith "))B matchin&.
Diabetes Reistry Data
Patient data $ere obtained by ;ueryin& the UCD3SCs
electronic medical record (%pic0 >e-rona0 @.). @e
included all patients $ith a dia&no-sis of DM $ho had
a clinic isit $ith a UCD3S family physician or
internist durin& the preious "-year period (7pril "50
2))50 to 7pril "*0 2))*). Confirmation of DM $as
based on .nternational Classification of Diseases
ersion * dia&nosis codes
28).)) throu&h 28).**. Patient ariables
re;uested included patient address0 a&e0 se!0
primary care proider name and office location0 last
&lycohemo-&lobin (7"c)0 last lo$-density
lipoprotein (DDD) cholesterol0 last urine
microalbumin<creatinine0 in-surance type0
race<ethnicity0 and primary lan&ua&e. The resultin&
table contained *#22 uni;ue patient records.
!""lication of Reistry Data to #$% Ma""in &ool
The diabetes re&istry data $ere incorporated into a
spreadsheet format (%!cel 2))-0 Microsoft0 2elle-ue0
@7). These data $ere conerted to a database file
structure for use in the /.S soft$are (7rc.nfo ersion
*.-0 %SE.0 Eedlands0 C7). /eocodin& $as performed
to obtain latitude and lon&itude coordi-nates for each
patientCs home address. .nitially0 5)8" (5-B) records
$ere matched and "'#" ("#B) $ere left unmatched.
,umerous errors $ere dis-coered durin& the
rematchin& process0 includin& abbreiated street
names0 lacA of appropriate spac-in& bet$een $ords
and<or street numbers0 and mis-spellin&s of street
names. 7ll unmatched records $ere reie$ed and
corrected $here possible0 lead-in& to another 4##
matched addresses (total of 5825 F55BG). Most of the
remainin& unmatched ad-dresses $ere post office
bo!es and therefore could not be &eocoded.
.ndiiduals youn&er than the a&e of 28 $ere e!cluded
(n 8-) so family physicians could be directly
compared $ith internists. .n ad-dition0 this a&e cutoff
correlates $ith the census bureauCs cutoff for
education attainment reportin&. 7nother ""8 patients
did not hae an 7"c on file and $ere e!cluded.
Eesultin& point features $ere then spatially ?oined to
year 2))) census tracts ($$$.census.&o). 7 spatial
?oin is a process that allo$s the /.S user to append
the attributes of one data layer (patient address
points) to the attributes of another layer (census data)
based on a common location (census tracts). 1or
cases in $hich a census tract had fe$er than ")
patients0 those tracts and patients (#'-) $ere
e!cluded from the analysis to aoid the rate instability
associated $ith small sam-ple siHes0 leain& #255
patient records for analysis (1i&ure ").
These data $ere used to construct 2 e!ample
maps as demonstrations of the capabilities of this
/.S tool. The first (1i&ure 2) addressed the dia-
betic population of the primary care net$orA+ the
second (1i&ure -) addressed the comparison of
doi9
").-"22<?a
bfm.2)").
)".)*)"4*
Usin& /.S to
7ssess Iutcome Disparities
5*
<igure 1. Study design showing initial patient data%ase and progressi'e exclusion criteria. Shaded %oxes
indicate num%ers of patients who were excluded.
driin& distance to
each patientCs primary
care pro-ider.
%ocioeconomic and
Demora"hic Data
Socioeconomic and
demo&raphic data $ere
ob-tained from
/eoDytics0 .nc. (%ast
2runs$icA0 ,(). @e
collected 2))5 data at
the census tract leel.
/eoDytics bases their
estimates on US
Census 2u-reau reports
and limited population
estimates0 then
e!pands on those to
proide multiple
population-based
ariables. >ariables
included $ere median
income0 education
attainment0
unemployment0 and
$hite and blacA race.
Data !nalysis
To address study
purposeJ describin&
the contrib-utin& factors
associated $ith
differences in out-
comes associated $ith
optimal &lucose and
lipid controlJ-$e
performed the follo$in&
data analy-sis. @e obtained patient data and
demo&raphic information from our disease
mana&ement re&istry. This aforementioned re&istry
is deried from our electronic medical records.
.nformation about so-
cioeconomic status
(S%S) $as obtained from
the US Census 2ureauCs
2))5 report. Patient0
socio-economic0 and
demo&raphic data $ere
?oined to a sin&le
database usin& 7rc.nfo
(ersion *.-0 %SE.).
Then0 %uclidean distance
$as calculated from each
patientCs home to their
primary care clinic. The
resultin& file $as loaded
into ST7T7 MP "" (Stata
Corp.0 Colle&e Station0
TK). Sociodemo&raphic
ariables $ere combined
usin& factor analysis. @e
follo$ed the techni;ue
described by DieH Eou!
et al
5
to create a
summary measure of
census tract leel
socioeconomic status.
>ariables in the factor
analysis included median
income in 2))50 propor-
tion of population $ith a
belo$-aera&e
education0 proporation of
the population that $as
unem-ployed0 and the
proportion of the
population that $as blacA
or $hite. The resultin&
factor (labeled as S%S)
$as cate&oriHed by
;uintiles.
@e sou&ht to
determine $hether
there $as an
association bet$een
optimal &lucose and
lipid con-trol $ith these
demo&raphic and
socioeconomic
ariables. .n our first
model $e conducted a
ran-dom-intercept and
random-slope
re&ression analy-
*) 361<=
(anuary=
1ebruary 2)") >ol. 2- ,o. "
http9<<$$$.?abfm.or&
Thank you for evaluating
BCL easyConverter Desktop
This @ord document $as conerted from PD1 $ith an ealuation
ersion of 2CD easyConerter DesAtop soft$are that only
converts the first 3 pages of your PD1.
7ctiate your soft$are for less than L2)
http9<<$$$.pdfonline.com<easyconerter<
CTEDM ClicA on the linA belo$ to purchase